sResearch partially supported by an NSF postdoctoral fellowship and by NSF grant DMS-9612317.
Topology Vol. 38, No. 4, pp. 823—859, 1999( 1999 Elsevier Science Ltd. All rights reserved
Printed in Great Britain0040-9383/99 $19.00#0.00
PII: S0040-9383(98)00030-5
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWSAND VIRTUAL FIBERS IN 3-MANIFOLDS
SERGIO R. FENLEYs
(Received 4 November 1996; accepted 12 May 1998)
1. INTRODUCTION
We study the topological behavior of immersed surfaces transverse to pseudo-Anosov flowsin closed 3-manifolds. Such flows induce singular stable and unstable foliations in thesurface. The main result is the following: we show that the surface is a virtual fiber if andonly if the induced foliations in the surface do not have closed leaves. The main tool isa careful study of lifts of the surface to the universal cover and how they relate to the lifts ofthe singular stable and unstable foliations. The topology of the induced foliations in thesurface is also described in detail.
A fundamental problem in 3-manifold theory is to classify 3-manifolds up to homeo-morphism. One of the simplest classes of 3-manifolds consists of those fibering over thecircle and they are tremendously common in three-dimensional topology. However givena 3-manifold it is in general very hard to decide whether it fibers over the circle or not, andthis is a very important question. Stalling [22, 33] gave a purely algebraic sufficientcondition (when the manifold is irreducible) for a manifold to fiber.
An equivalent question is to decide whether a given surface in the manifold is a fiber ofa fibration of the manifold over the circle. In general, it is better to consider the followingmore general setting: a surface (not necessarily embedded) is called a virtual fiber if there isa finite cover of the manifold so that the surface lifts to a surface homotopic to a fiber in thiscover. Again one has a question which is very easy to state but hard to solve: given a surfacehow does one tell whether it is a virtual fiber or not?
There is a general characterization of virtual fibers, also due to Stallings [33]: supposethe manifold is irreducible and the surface is incompressible, where incompressible meansinjective in the fundamental group level. Then the surface is a virtual fiber if and only itsfundamental group is a virtual normal subgroup of the fundamental group of the manifold[22] (in this article the term incompressible is also used for non embedded surfaces). Thereare also more technical characterizations using properties of systems of curves in thesurface [23, 39].
A general answer to the virtual fiber question was also obtained when the manifold ishyperbolic. Notice that the class of hyperbolic 3-manifolds is quite large [38]. The universalcover of such a manifold is hyperbolic 3-space which is compactified with a sphere atinfinity. When the surface is a virtual fiber then the limit set of any lift SI of the surface S to
823
the universal cover is the entire sphere at infinity [35]. Another possibility is that the surfaceis quasi-Fuchsian, that is, the group of covering translations associated to the surface isquasi-conformally conjugate to a Fuchsian group [26]. In this case the limit set of SI isa Jordan curve. The fundamental result obtained by combining results of Marden, Thurstonand Bonahon states that these two diametrically opposite behaviors are the only twopossibilities:
THEOREM ([4, 26, 35]). ¸et S be an incompressible surface in M3 closed hyperbolic. ¹heneither
(i) S is quasi-Fuchsian or equivalently the limit set of SI is a Jordan curve.(ii) S is a virtual fiber or equivalently the limit set of SI is the entire sphere.
Notice that this characterization does not depend on intrinsic objects, like the image ofthe imersion in M; rather it relies on information about limit sets in the sphere at infinity (anextrinsic object), or the study of the large-scale geometry of SI .
Cooper et al. [7] studied the virtual fiber question for incompressible, immersed surfacesin hyperbolic 3-manifolds which fiber over the circle. They considered surfaces transverse tothe suspension flow of a pseudo-Anosov homeomorphism of a closed hyperbolic surface[36, 37]. Such flows have singular stable and unstable two-dimensional foliationsF4, F6 inM, which induce singular stable and unstable foliations F4
S, F6
Sin S. In a beautiful paper
[7] they completely determined the geometric behavior of S once the topological structureof F4
S, F6
Sis known.
THEOREM (Cooper et al. [7]). ¸et S be an incompressible, immersed surface transverse toa suspension of a pseudo-Anosov homeomorphism of a hyperbolic surface. ¸et F4
S, F6
Sbe the
induced singular foliations in S. ¹hen S is a virtual fiber if and only if F4S
has no closed leaves.Furthermore, if F4
Shas a closed leaf, then an arbitrary leaf of F4
Sis either closed or each of its
rays limits to a closed leaf of F 4S. Similarly for F6
S.
Since the manifold fibers over the circle, it is easy to construct many examples ofimmersed surfaces transverse to the flow and which are not homotopic to embeddedsurfaces. This provides many test cases for the virtual fiber question [7]. They extendedthese results in [8] to produce surprising examples of immersed surfaces which are nothomotopic to fibers but which are virtual fibers.
The idea of considering surfaces transverse to flows has been used in various othercontexts in 3-manifold theory and is interesting because the dimensions of flow/surface arecomplementary. In the present situation understanding the relative position (that is thetopology of the induced foliations) gives information about the topological situation in themanifold. This is part of a more general philosophy: good position of surfaces with respectto essential laminations (or foliations) gives information about the topology of the surfaceand/or the manifold, see for example [27, 28].
The goal of this article is to extend the above result to surfaces transverse to a muchmore general class of flows, namely pseudo-Anosov flows. A pseudo-Anosov flow is a flowthat is transversely hyperbolic (Anosov behavior [1]) almost everywhere, except that it mayhave finitely many singular orbits where it has p*3 prong singularities (see definition inSection 2). Anosov flows are exactly those pseudo-Anosov flows which do not have singularorbits. Suspensions of pseudo-Anosov homeomorphisms of closed surfaces form anotherclass of examples. Notice that pseudo-Anosov flows are quite abundant: Mosher showedthat every closed, hyperbolic 3-manifold with non zero second homology has many
824 S. R. Fenley
pseudo-Anosov flows [29]. In addition, pseudo-Anosov flows survive after almost all Dehnsurgeries along a given closed orbit of the flow [16] (except for the meridional surgery). Inparticular, there are many pseudo-Anosov flows which are not topologically conjugate tosuspension flows and also there are many examples where the underlying manifolds are nothyperbolic.
Pseudo-Anosov flows have singular two-dimensional stable F4 and unstable F6 foli-ations. The singular orbits of the flow are contained in singular leaves of the foliation. Nowconsider S, an immersed surface transverse to a pseudo-Anosov flow ' in a closed3-manifold M. By transversality of S and ', the two-dimensional singular foliationsF4, F6
induce one-dimensional singular foliations F4S, F6
Sin S. The first result shows that incom-
pressibility of S follows from the general setting of ' being pseudo-Anosov and S transverseto '. Notice that incompressibility was an additional hypothesis in [7], which then had tobe checked for all examples they analysed. In this article we will not assume that M ishyperbolic or that ' is a suspension.
THEOREM A. ¸et ' be a pseudo-Anosov flow in M3 closed. ¸et S be an immersed surfacetransverse to '. ¹hen S is incompressible. If oJ :SI PMI is a lift of S to the universal cover ofM then oJ (SI ) is a properly embedded plane in MI .
This theorem is proved by first observing that F4S
is a singular foliation in S. It is wellknown that up to Reeb annuli, such foliations are essential [24]. This gives informationabout how the surface sits in the 3-manifold and in the universal cover also. Next we analysethe virtual fiber question and prove:
MAIN THEOREM. ¸et S be an immersed surface in closed M3 so that S is transverse toa pseudo-Anosov flow ' in M. ¸et Fs
Sbe the induced singular stable foliation in S. ¹hen S
is a virtual fiber if and only if F4S
has no closed leaves. Furthermore, if S is a virtual fiberthen ' is topologically conjugate to a suspension of a pseudo-Anosov homeomorphism ofa closed surface and every leaf of F4
Sis dense in S. Finally if F4
Shas closed leaves then it has
finitely many closed leaves and every ray of F4S
limits to one of these closed leaves. Similarlyfor F6
S.
This completely generalizes the results in [7] to general pseudo-Anosov flows. WhenF4
Shas closed leaves, Cooper et al. called F4
Sa finite foliation. In the context of foliation
theory this is also called a depth one foliation [17]. These occur for instance in [9] whichstudies examples of flows transverse to two-dimensional, depth one foliations in hyperbolic3-manifolds.
Here are two remarks concerning the applications of the main theorem. If there isa virtual fiber transverse to ', then the theorem states that M fibers over the circle and ' isessentially a suspension flow. Hence, if one is interested in constructing virtual fibers whichare not homotopic to fibers as was one of the goals in [7, 8] then one only needs to considersurfaces transverse to suspension flows. If on the other hand a surface S transverseto a pseudo-Anosov flow is given, then the topology of F6
S, F4
Seither gives essential
information about the topology of M (fibration case) or it follows that F4S, F6
Sare finite
foliations.In the situation analysed in [7], M was hyperbolic and S incompressible, which forced
S to be a hyperbolic surface. We do not have that restriction, for instance S may be the fibertransverse to a suspension Anosov flow, in which case S is an euclidean surface. Elliptic
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 825
surfaces are disallowed by Euler characteristic reasons. Our results include the case that S iseuclidean, where most proofs are simpler. Notice that there may be examples of toritransverse to (singular) pseudo-Anosov flows, which could happen if they do not intersectthe singular orbits.
The main theorem concerns homotopic properties of S and as hinted by theorem A, it isessential to understand the picture in the universal cover. It turns out that there is a simplecharacterization of virtual fibers when lifting to the universal cover:
THEOREM B. ¸et S be an immersed surface transverse to a pseudo-Anosov flow ' inM3 closed. ¸et '3 be the lift of ' to MI and oJ (SI ) a lift of S to MI . ¹hen S is a virtual fiber if andonly if oJ (SI ) intersects every orbit of '3 .
We use this theorem as an intermediate step to prove the main theorem. In the universalcover the global structure of the (two-dimensional) singular foliations is relatively simpleand it is quite useful for our purposes. The two possibilities for the behavior of lifts of S toMI can then be related to the topology of F4
Sand F6
Sin the manifold. Given theorem B, one
direction of the equivalence is easy (if there are closed leaves inF4Sthen S cannot be a virtual
fiber). The other direction is quite hard and relies heavily on the topological structure of thestable and unstable foliations in the universal cover. The strategy for the proof of the maintheorem is to first show that the only two possibilities for F4
Sare either minimal foliation or
finite foliation and then associate these options to being a virtual fiber or not.The idea of first relating to the universal cover and then back to the singular foliations in
S was also used in [7], who proved the same result as Theorem B, using it as an intermediatestep. However the tools used here are completely different than those of [7] and we nowexplain why.
The results in [7] depend heavily on the following geometric data: (1) the manifold ishyperbolic, where for example one can apply the fundamental dichotomy for surfaces asrelated to the behavior of limit sets of lifts to MI ; (2) the flow, being a suspension flow, istherefore quasigeodesic. Quasigeodesic means that flow lines are uniformly efficient inmeasuring distances in relative homotopy classes. In our general situation we do not havethese geometric properties, but we can show the same theorem using only the dynamicalproperties of the flow and the foliations associated to it. We will only use the geometry whenit is deduced that the manifold fibers over the circle and the flow is a suspension, in whichcase there is a lot of geometric information.
One important fact which will be used throughout the article is that '3 is a product flow:if O is the orbit space of '3 obtained by collapsing orbits of '3 to points, then O is Hausdorffand homeomorphic to R2 [15]. This easily shows that M is irreducible and MI is homeomor-phic to R3. When ' is the suspension flow of a pseudo-Anosov homeomorphism of a closedsurface, the set O may be naturally identified to the universal cover FI of a fiber F ofa fibration of M over the circle. Then O:FI is quasi-isometric to the hyperbolic planeH2 and has a well defined circle at infinity S1
=. In addition, covering translations of
MI project to homeomorphisms of O:FI , which act metrically in O (as lifts of the pseudo-Anosov homeomorphism of the closed surface). These geometric properties were alsoessential in [7]. The problem is that for general pseudo-Anosov flows there is no naturalmetric in O — this set is only a topological object. Still one has the product topologicalstructure of the flow and the singular foliations in MI , so we can use the topological theory ofpseudo-Anosov flows. The additional lucky tool is the fact that singular foliations insurfaces have very good properties.
As a corollary of our analysis we obtain the following result:
826 S. R. Fenley
COROLLARY C. ¸et S be an immersed surface, transverse to a pseudo-Anosov flow inM3 closed hyperbolic. ¹hen S is quasi-Fuchsian if and only if the induced singularstable/unstable foliations in S have closed leaves. Otherwise S is a virtual fiber.
Finally, we mention that Brian Mangum [25] has previously obtained, using differentmethods, a proof of Theorem A and a partial proof of Corollary C, both in the case that themanifold is hyperbolic, the flow is quasigeodesic and has no dynamic parallel orbits asdefined by Mosher in [27, 28]. These include suspension flows.
The article is organized as follows: the next section contains the necessary backgroundmaterial. In Section 3 we show that immersed surfaces transverse to pseudo-Anosov flowsare always incompressible and in the next section we prove Theorem B. The last section isthe most technical, where we use the characterization of Theorem B to study the singularfoliations in S and prove the main theorem.
2. PSEUDO-ANOSOV FLOWS AND SINGULAR FOLIATIONS ON SURFACES
Pseudo-Anosov flows are a generalization of suspension flows of pseudo-Anosov surfacehomeomorphisms. These flows behave much like Anosov flows, but they may have finitelymany singular orbits which are periodic and have a prescribed behavior. In order to definepseudo-Anosov flows, first we recall singularities of pseudo-Anosov surface homeomorphisms.
Given n*2, the quadratic differential zn~2 dz2 on the complex plane C (see [34] forquadratic differentials) has a horizontal singular foliation f 6 with transverse measure k6, anda vertical singular foliation f 4 with transverse measure k4. These foliations have n-prongedsingularities at the origin, and are regular and transverse to each other at every other pointof C. Given j'1, there is a homeomorphism t : CPC which takes f 6 and f 4 to themselves,preserving the singular leaves, stretching the leaves of f 6 and compressing the leaves of f 4 bythe factor j. Let Rh be the homeomorphism zPe2n*hz of C. If 0)k(n the map R
k@n°t has
a unique fixed point at the origin, and this defines the local model for a pseudohyperbolicfixed point, with n-prongs and rotation k. This map is everywhere smooth except at theorigin. Let dE be the singular Euclidean metric on C associated to the quadratic differentialzn~2 dz2, given by
d2E"k2u#k2
s.
Note that(R
k@n°t)* d2E"j~2k2
u#j2k2
s.
The mapping torus N"C]R/(z, r#1)&(Rk@n°
t(z), r) has a suspension flow ( arisingfrom the flow in the R direction on C]R. The suspension of the origin defines a periodicorbit c in N, and we say that (N, c) is the local model for a pseudohyperbolic periodic orbit,with n prongs and with rotation k. The suspension of the foliations f 6, f 4 define two-dimensional foliations on N, singular along c, called the local weak unstable and stablefoliations in a neighborhood of c. In addition, the sets which are of the form (leaf of f 4)]MtN,where t3S1, form a one-dimensional singular foliation in N. The singular locus is exactly c.This foliation is the local strong stable foliation. Similarly the sets (leaf of f 6)]MtN define thelocal strong unstable foliation.
Note that there is a singular Riemannian metric ds on C]R that is preserved by thegluing homeomorphism (z, r#1)&(R
k@n"t(z), r), given by the formula
ds2"j~2tk2u#j2tk2
s#dt4
The metric ds descends to a metric on N denoted dsN.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 827
Definition 2.1. Let ' be a flow on a closed, oriented 3-manifold M. We say that ' isa pseudo-Anosov flow if the following are satisfied:
f For each x3M, the flow line tP't(x) is C1. The tangent vector bundle D
t' is C0
in M.f There is a finite number of periodic orbits Mc
iN (this may be an empty set), called
singular orbits, such that the flow is smooth off of the singular orbits.f Each singular orbit c
iis locally modelled on a pseudo-hyperbolic periodic orbit. More
precisely, there exist n, k with n*3 and 0)k(n, such that if (N, c) is the local modelfor pseudo-hyperbolic periodic orbit with n prongs and with rotation k, then there areneighborhoods º of c in N and º
iof c
iin M, and a diffeomorphism p :ºPº
i, such
that p takes orbits of the semiflow Rk@n°
t D º to orbits of ' D ºi.
f There exists a path metric dM
on M, such that dM
is a smooth Riemannian metric offthe singular orbits, and for a neighborhood º
iof a singular orbit c
ias above, the
derivative of the map p : (º!c)P(ºi!c
i) has bounded norm, where the norm is
measured using the metrics dsN
on º and dM
on ºi.
f On M!Zci, there is a continuous splitting of the tangent bundle into three
one-dimensional line bundles E6=E4=¹', each invariant under D', such that ¹' istangent to flow lines, and for some constants k
0'1, k
1'0 we have
1. ! If v3E6 then ED't(v)E)k
0ek1tEvE for t(0,
2. ! If v3E4 then ED't(v)E)k
0e~k1tEvE for t'0,
where norms of tangent vectors are measured using the metric dM
.f The map p~1 which translates back the neighborhoods (º
i, c
i) to the canonical local
model (º, c) by p~1 satisfies: the vectors dp~1(E4) and dp~1(E6) are tangent to thestrong stable and unstable foliations in º!c .
With this definition, pseudo-Anosov flows are a generalization of Anosov flows in3-manifolds [1]. The entire theory of Anosov flows can be mimicked for pseudo-Anosovflows. In particular, a pseudo-Anosov flow ' has a singular two-dimensional weak unstablefoliation F6 which is tangent to E6=¹' away from the singular orbits, and a two-dimensional weak stable foliation F4 tangent to E 4=¹'. These foliations are singularalong the singular orbits of ', and regular everywhere else. In the neighborhood º
iof an
n-pronged singular orbit ci, the images of F4 and F6 in the model manifold N are identical
with the local weak stable and unstable foliations.The pseudo-Anosov flow also has singular one-dimensional strong foliations F44, F66.
Leaves of F44 are obtained by integrating E4 away from the singularites and patchingtogether with the singular one dimensional strong stable foliation defined in a neighbor-hood of a singular orbit, using the final condition of the definition. The foliation F44 is flowinvariant, that is, for any leaf q of F44 and any real t, '
t(q) is a leaf of F44. Furthermore, for
t'0 the time t homeomorphism 'texponentially contracts distances along leaves of F44.
Similarly for F66.This discussion applies equally well to the lifted singular foliations FI 4, FI 6, FI 44, FI 66 in
MI . If p3M let ¼4 (p) denote the leaf of F4 containing p. Similarly define ¼6(p), ¼44(p),¼66(p) and in the universal cover ¼I 4(p), ¼I 6 (p), ¼I 44(p), ¼I 66(p).
We now discuss geometric properties of singular foliations on surfaces. Most factsdescribed here are in some form mentioned in [35, 37]. Explicit proofs were done by Levittin [24].
Let F be a singular foliation in a closed hyperbolic surface S so that each singularity isa p-prong singularity with p*3 prongs. A Reeb component in dimension 2 (here called
828 S. R. Fenley
a Reeb annulus) is an annulus RLS, which is saturated by F and satisfies: each componentof LR is a closed leaf of F and in the interior of R there are no singularities, all leaves arenon-compact, the set of such leaves forms a fibration over the circle, and finally each leafspirals towards LR in both directions, so that there is no transversal to F going from onecomponent of LR to the other.
The surface S is assumed to have a fixed hyperbolic metric. Associated to F there isa unique geodesic lamination G obtained as follows. There are at most finitely many Reebannuli in F. Removing the interior of these annuli produces a Reebless foliation F @ ofa closed subset of S. Next we pull tight the leaves of F @. Let FI @ be the lift of F @ to theuniversal cover SI . Recall that SI is isometric to the hyperbolic plane H2 which is canonicallycompactified with a circle at infinity S1
=.
We define a map gJ from the leaves of FI @ to a collection of geodesics of H2 by: Levitt [24]proved that if l is a regular leaf of FI @ (that is l contains no singularities) then l has twowell-defined distinct ideal points in LH2"S1
=and gJ (l) is the geodesic defined by these ideal
points. The surface foliations we will be interested in this article will have at most onesingularity per leaf. In that situation, if a leaf l of F@ has a p-prong singularity, then split thesingularity in l to produce p non singular leaves and subsequently pull each of those leavestight to produce p geodesics in SI . This is gJ (l).
The geodesic lamination GI thus produced in SI is n1(S) invariant, generating a geodesic
lamination G in S. This is the geodesic lamination associated to F and is denoted byG"g(F). One obtains an associated map g from leaves of F@ to leaves of G. It ismultivalued in the singular leaves of F@.
A leaf m of GI is thick if there are two distinct leaves h, h@ ofFI @ with gJ (h)"gJ (h@)"m. Thisimplies that h and h@ are disjoint and in SI they bound a connected open set º(h, h@) so thatany leaf l of FI @ contained in º (h, h@) satisfies gJ (l )"m and any two such leaves are disjoint.Consequently, there are extreme leaves h0,h1 so that given any leaf l ofFI @ with gJ (l )"m thenl belongs to h0Xh1Xº(h0, h1). If there is a singularity in h
0then we usually denote by
h0
only the piece l of h0
which after spliting the singularity would satisfy gJ (l)"m. The seth0Xh1Xº(h0, h1) is the band of FI associated to m. There are no singularities of FI @ inº(h0, h1) (there may be singularities in the boundary). The set of thick geodesics is preservedby the action of n
1(S) and projects to a family e (F) of thick geodesics of G.
Let g be a thick geodesic of GI . If projection of the associated band½"h0Xh1Xº(h0, h1) does not embedd in S, then there is covering translation f of SI withf (½ )W½O0. Since ½ is maximal with respect to being a band, this implies that f leavesinvariant the boundary leaves of ½. Therefore ½ projects to an embedded annulus orMoebius band in S with the boundaries being closed leaves of F. We conclude that if theprojection of ½ in S does not contain a closed leaf ofF, then½ is mapped injectively into S.
At this point we should discuss half Reeb components which are the following: let Z behomeomorphic to (0, 1]]R with a foliation by leaves MxN]R. Now consider Z as part ofa singular foliated 2-manifold, where M1N]R may have a prong singularity with the otherprongs going to the exterior side of Z. In addition, we require that the lift of Z to theuniversal cover of the surface is properly embedded. Even though this situation may welloccur, for instance for singular (or even non-singular) foliations of the plane, this does notoccur for foliations of compact surfaces. Here is why: Consider the rays a
1,a
2of M1N]R.
Since S is compact, these rays limit somewhere in S. If the distance from a1
to a2
does notconverge to 0 along the rays, then it follows that Z is not embedded in S and in fact thisproduces a Reeb annulus. If the distance between a
1and a
2converges to 0, then they
eventually are in the same foliation box. Close up this long segment in the ray with a shorttransversal to produce a closed curve bounding a disc D in S. Index computations of
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 829
singularities in D show that there must be either a 1-prong singularity or a center in D, bothnot allowed here. We conclude that half Reeb components do not exist in our setting.
One can now understand general position of curves in S with respect to F. Firstconsider the case that F does not have Reeb annuli. Since the associated lamination G isa geodesic lamination, it follows that each essential closed curve c in S is freely homotopic toa curve transverse to G. This implies that c is freely homotopic to a curve which is eithercontained in a leaf of F or is transverse to F, where one allows the curve to pass throughsingularities of F. We remark that if there were more than one singularity in a leaf, then thebest representative of c could be forced to contain a segment in a leaf of F.
In the presence of Reeb components, this shows that any essential closed curve c in S isfreely homotopic to a curve c@ which is transverse to F@. In addition, c@ may be chosen sothat in each Reeb annulus E of F, c@ consists of arcs from one component of LE to the other,each arc being tangent to F in a single point. The curves with good intersection with F arecalled efficient representatives of c with respect to F.
If, for instance, a Reeb annulus A has a singularity in a boundary A1
of A here is whathappens: In the universal cover, AI
1would have infinitely many singularities and the
splitting procedure described above would produce infinitely many leaves, hence infinitelymany geodesics in GI . Still those geodesics are invariant under g, where g is the coveringtranslation associated to A
1. This produces only finitely many geodesics of G. In the next
section we show that for F4S, F6
S, the boundaries of the Reeb annuli do not have singular-
ities, thus this situation does not occur.Remark 3.1 of [24] shows that if a sequence of leaves l
nof FI @ converges to l in FI @, then
gJ (ln) converges to gJ (l ). This implies that the leaf space of FI @ is Hausdorff. Otherwise we find
hnPhXh@, hOh@. But then gJ (l
n)PgJ (l )XgJ (l@ ) and since the leaf space of a geodesic lamina-
tion in H2 is Hausdorff, it follows that gJ (l)"gJ (l@ ), contradiction. Hence the leaf space of FI @is Hausdorff. Notice that this leaf space is not a 1-manifold, but rather it is a one-dimensional tree (this is in the case F@ is a foliation, that is, there are no Reeb annuli).
A complementary region E of FI @ is the lift of a Reeb annulus of F and is bounded bytwo curves a
1, a
2which project to closed curves in S. Then a
1and a
2have the same ideal
points in S1=. Since distinct leaves of FI @ are separated, this shows that if lOl@3FI are not
separated from each other in FI , then l, l@ are in the boundary of one such complementaryregion E. This implies that the projections of l and l@ in S are closed and bound a Reebannulus.
Now suppose that S is a closed Euclidean surface. The restriction on the type ofsingularities (p-prong with p*3) implies that in fact there are no singularities in F. Asbefore there may be finitely many Reeb annuli which in this case are all isotopic. If there areno Reeb annuli, the foliation is conjugated to a suspension of a homeomorphism of thecircle [21]. Hence, all leaves can be pulled tight to geodesics and exactly the same results asbefore are obtained as for hyperbolic surfaces. Similarly when there are Reeb annuli and wealso get the same results as for hyperbolic surfaces.
3. HOMOTOPIC PROPERTIES OF THE SURFACE
In this section we show that immersed surfaces transverse to pseudo-Anosov flows arealways incompressible. We also show that any lift of the immersion o :SPM to theuniversal cover oJ : SI PMI is an embedding. The idea is to represent any homotopically nontrivial closed curve in S by an efficient curve with respect to F4
Sand then show that it is not
null homotopic in M using the foliations F4, F6. We will sometimes denote by S both thesurface in its intrisic topology and its image in M. When it is important to differentiate
830 S. R. Fenley
between them we write S and o (S). We remark that many proofs work equally well forS hyperbolic or Euclidean. We will not differentiate unless needed.
In the orientable double cover of M, S lifts to an immersed surface transverse toa non-singular flow, hence orientable. Since almost all results in this article are invariantunder finite covers, we assume from now on that M is orientable unless otherwise stated.
Let '3 be the lift of ' to MI . Let O"MI /'3 be the orbit space. A fundamental fact for us isthat O is homeomorphic to the plane (O+R2) [15], that is, '3 is a product flow in MI . ThenFI 4, FI 6 induce transverse singular foliations in O, denoted by FI 4O, FI 6O. Let # : MI PO be theprojection map. Notice that there is no natural metric in O — it is only a topological object.
The following simple result will be used throughout the article.
LEMMA 3.1. A an arbitrary leaf of F4Scan have at most one singularity and a closed leaf of
F4Sdoes not have any singularity. If FI 4
Sis the lift of F4
Sto SI , then a leaf of FI 4
Scan have at most
one singularity. Similarly for F6S.
Proof. Suppose that a leaf l ofF4Shas two singularities p and q. Lift to lI inFI 4
Sto produce
two singularities pJ , qJ in lI . Similarly, if a closed leaf l has a singularity at p, then in any lift lI toSI , there are infinitely many distinct lifts of p, hence infinitely many singularities of FI 4
S.
Therefore, it suffices to show that leaves of FI 4S
have at most one singularity.
Suppose by contradiction that a leaf r of FI 4Shas two singularities a and b and let c be the
compact segment in r between a and b. Fix a lift oJ :SI PMI of S to MI . Here it is convenient todistinguish between FI 4
Sand oJ (FI 4
S). Let ¸3FI 4 with oJ (r)L¸. The leaf ¸ is singular and has
a unique singular orbit c. Notice that '3 is a product flow in ¸"¼I 4 (c ), that is, ¼I 4(c) ishomeomorphic to # (¼I 4 (c))]R (where #(¼I 4(c)) is an n-pronged leaf ofFI 4O in the plane O)and the foliation by flow lines of '3 in ¼I 6 (c) is the vertical foliation. Since oJ (r) is an arc in¸ with endpoints oJ (a), oJ (b), both of which are in the singular orbit c, it follows from theproduct picture in ¼I 4 (c) that there is at least one point in oJ (r) where oJ (r) is not transverseto '3 . This contradicts the fact that oJ (SI ) is transverse to '3 and finishes the proof ofthe lemma. K
This has important consequences. Let a be a closed curve in S. As described in theprevious section g(F4
S) is a geodesic lamination in S and a can be freely homotoped to have
efficient intersection with g (F4S). When collapsing to the foliations setting the following
happens:
(1) Since there is at most one singularity in a leaf of F4S, it follows that the efficient
representative a1
does not contain a segment in a leaf of F4S, unless a
1is a leaf of F4
S.
(2) If a1
contains a singularity v of F4S, then a
1crosses the singularity. This means that
the leaf of F4S
through v has at least 4 prongs at v and locally a1
separates at least twoprongs on each side.
(3) If b is a boundary leaf of a Reeb annulus, then b is a closed leaf and thereforecontains no singularity of F4
S.
Definition 3.2. Given a leaf ¸ of FI 4 or FI 6, let stab(¸) be the stabilizer of ¸ in n1(M).
Notice that stab(¸) is either trivial or infinite cyclic.
THEOREM 3.3. ¸et S be an immersed surface transverse to a pseudo-Anosov flow inM3 closed. ¹hen S is incompressible. If oJ :SI PMI is a lift of S to MI , then oJ (SI ) is a properlyembedded plane. Finally oJ (SI ) intersects every orbit of '3 at most once, that is, #DoJ (SI ) is injective.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 831
Proof. For this result it does not matter whether S is euclidean or hyperbolic. We willconsider efficient representatives of closed curves in S and analyze how they behave inM with respect to F4.
The first task is the following : let D be a Reeb annulus of F4Sand DI be a lift to MI . We
need to understand the relative position of DI with respect to FI 4.Fix a lift oJ : SI PMI . Let E be a Reeb annulus in F4
Swith boundaries E
1,E
2and interior
leaves Gt, t3S1. Then E
1is a closed curve in a leaf F of F4 and E
1is transverse to '. Using
cut and paste surgery on E1
in F, one produces a collection of simple closed curves inF which are transverse to '. By Euler characteristic reasons, these curves cannot be nullhomotopic in F and hence F is not simply connected and contains a periodic orbit c
1of '.
Similarly, E2LF
23F4 and c
2is the periodic orbit of ' in F
2. Now lift E to EI LSI and with
boundaries EI1, EI
2. Consider oJ (EI )LMI . Let EI
iL¸
i3FI 4. Since E is a Reeb annulus, then
EI1, EI
2are not separated from each other in the leaf space of FI 4
S. This implies that ¸
1is not
separated from ¸2in FI 4 in the leaf space of FI 4. The key fact needed to prove the theorem is
that these are different leaves of FI 4.
LEMMA 3.4. ¹he leaves ¸1, ¸
2are distinct, ¸
1O¸
2.
Proof. Suppose not, that is, ¸1"¸
2. Let Mc be the lift of M associated to the infinite
cyclic subgroup stab(¸1)Ln
1(M). Since ¸
1"¸
2they project to the same stable leaf in Mc.
We denote this by ¸c. Let c be the periodic orbit in this leaf. The cover McPM projects¸c homeomorphically onto its image.
Let nc : MI PMc be the covering projection and 'c be the induced flow in Mc.We refer to Gabai and Oertel [18] for definitions and details concerning essential
laminations.
CLAIM. ¹he leaf ¸c is properly embedded in Mc.
In order to show the claim, first blow up each singular leaf of F4 in M to produce a non-singular lamination V4 in M. This lamination satisfies the following properties:
(1) V4 has leaves which are either planes, annuli or Moebius bands,(2) the complementary regions of V4 are homeomorphic to open solid tori or solid
Klein bottles of the form (ideal n-gon) ][0,1], so that top and bottom are glued bya homeomorphism. These are called twisted components. This term will be usedthroughout the article. The definition of pseudo-Anosov flows implies that n*3.
It follows from [18] that V4 is an essential lamination. We remark that this shows thatleaves of FI 4, FI 6 are properly embedded and separate MI [18].
Lift V4 to a lamination V4c in Mc. Suppose for simplicity that c is not a singular orbitand hence we may assume that ¸c is a single leaf of V4c. If ¸c limits on p3Mc, then takea very long path in ¸c with endpoints q, z near p, connect its endpoints by a small arctransverse to V4c and finally perturb it slightly to produce a transverse curve b to V4c . If it isnull homotopic in Mc, then it projects to a null homotopic transversal to V4, contradictionto V4 being an essential lamination [18].
If b is not null homotopic, then consider n1(Mc) with basepoint q. Since n
1(Mc) is
generated by [c], it follows that b is freely homotopic to cn for some n3Z!M0N. Lift b tobI in MI starting in qJ 3¸
1. Since ¸c is invariant under stab(¸c)"n
1(Mc), then the other
endpoint of bI is in ¸c also. Therefore bI can be closed up by a segment in ¸c and the resulting
832 S. R. Fenley
closed loop bI @ can again be slightly perturbed to be transverse to FI 4. As nc (bI @) is nullhomotopic in Mc, this produces a contradiction as before. This finishes the proof of the claim.
Continuation of the Proof of ¸emma 3.4. At this point we return to the foliation setting(F4 ) as opposed to using the lamination setting (V4).
If ¸c is a Moebius band, lift to a double cover and assume from now on that ¸c is anannulus. Let Ec"nc(EI ) which is an immersed annulus in Mc. Let LEI be the union of theboundary components of EI in MI and let
int (EI )"EI !LEI , int (Ec)"nc (int(EI )) and LEc"nc(LEI ).
Notice that LEcL¸c. If ¸c intersects int(Ec) in a point z, then because Ec is (intrinsically)a Reeb annulus, it follows that the leaf of int(Ec)W¸c through z limits on points of¸c (namely those in LEc). In that case ¸c would not be properly embedded in Mc,contradicting the above claim.
The important consequence is that int(Ec)W¸c"0. Now we are ready to do cut andpaste surgery to Ec producing a union of embedded surfaces BcLMc which are stilltransverse to 'c. This cut and paste surgery transverse to flows was introduced by Fried in[16]. The fact int (Ec)WLEc"0, implies that the surgery is only done along closed curvesand arcs so that the boundary points of the arcs are not in int(Ec). It follows that the totalEuler characteristic of Ec is unchanged by the cut and paste surgery and no prongsingularities are created in the boundary of Bc. In general, if int(Ec)WLEcO0, then cut andpaste will create prong singularities in the boundary and the Euler characteristic willchange. This was the reason for proving that ¸c is properly embedded in Mc.
Therefore, the total Euler characteristic of Bc is the same as that of Ec , which is zero. Inaddition there are components of Bc with boundary. Let Cc be one such component withboundary. Since Mc is orientable and Ec is transverse to the flow, it follows that the cut andpaste yields orientable surfaces. There are no centers created and LCc is contained in a unionof leaves of the stable foliation of Mc , therefore there is no sphere or disk component in Bc .As the total Euler characteristic is unchanged, it follows that there are no components withnegative Euler characteristic in Bc , so all components of Bc have zero Euler characteristic.We conclude that Cc is an annulus.
Furthermore LCcL¸c is a union of 2 distinct embedded curves C1,C
2transverse to
'c in ¸c. It follows that C1,C
2are freely homotopic to c or c~1 in ¸c and C
1XC
2bounds an
annulus Dc in ¸c (they cannot bound a Moebius band because ¸c is an annulus). Let¹"CcXDc, which is embedded and either a torus or a Klein bottle. The case ¹ is a Kleinbottle would imply that Cc is a free homotopy from c to c~1 in Mc . This contradicts the factthat n
1(Mc)+Z and [c] is a generator. Therefore ¹ is a torus.
Since n1(¹) is Z=Z it follows that it is not injective in n
1(Mc)"Z. As ¹ is embedded
there is a simple closed curve in ¹ bounding an embedded disk D1in Mc with D
1W¹"LD
1and LD
1not null homotopic in ¹ [22]. Cut along D
1to produce a sphere in Mc. Since MI is
homeomorphic to R3, this sphere bounds a ball in Mc . There are two options: it mayhappen that one component of Mc!¹ is contained in this 3 ball — contradiction becausea component of LCc is freely homotopic to c in Mc and cnOid in n
1(Mc), for any nO0. The
remaining option is that ¹ bounds a solid torus ».The flow 'c is transverse to CcL¹ and it is tangent to Dc . If the flow is outgoing
» along Cc then in ¸c the flow 'c would be transverse and outgoing in the two componentsof LDc . This contradicts the fact that ¸c is a stable leaf. It follows that 'c is transverse toCc going into » and also that c is contained in the interior of Dc . Because no flow lines aretransverse to ¹ going out of » and cL¹, it follows that a component º of ¼6c(c)!c is
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 833
contained in int(»). Lift to the universal cover. Then º lifts to ºI contained in the infinitesolid cylinder int(»I ). Consider points x
iin ºI arbitrarily far from cJ in the path metric of ºI .
Since »I is the lift of a compact solid torus and »I is invariant under [c], we may assume thatall x
iare in a compact subset of MI . This contradicts the fact that ¼I 6(c) is properly
embedded in MI [18]. This final contradiction implies that ¸1O¸
2and hence finishes the
proof of Lemma 3.4. K
Continuation of the proof of ¹heorem 3.3. Let now d : [0, 1]PS, with d(0)"d(1) be anessential closed curve in S. We can assume that it has efficient intersection with F4
Sas
described in the previous section. Assume for simplicity that d(0) is not in the interior ofa Reeb annulus of F4
Sand that d(0) is not a singularity of F4
S. Suppose first that d is not
contained in a leaf of F4S. Lift d to dI in SI and consider oJ °dI in MI starting at q and ending in
p and let
Z"Mt3[0, 1] Dd(t) is not in the interior of a Reeb componentN.
Then Z"[0, 1] — finite union of open intervals. Let Ftbe the leaf of FI 4 containing
oJ (dI (t)).
CLAIM. For any t, r, s3Z with t(r(s, then Frseparates F
tfrom F
s.
Notice that there are at most finitely many Reeb annuli in F4S. If d(0) is not in the
boundary of a Reeb annulus this implies that there is e'0 so that d is transverse to F4Sin
[0, e] and [0, e]LZ. Lifting to the universal cover, one sees that the claim holds fort(r(s)e.
(i) Let t0be the first time d (t) is in a Reeb annulus and assume that the claim holds for all
t(r(s)t0
which are in Z. Let t1
be the time d(t) exits this first Reeb annulus. ThenFt0, F
t1are not separated in FI 4 and the previous lemma shows that F
t0OF
t1. By construc-
tion Ft0
separates Ftfrom F
t1if t(t
0(see Fig. 1(a)), so the claim holds for all t(r(s)t
1which are in Z.
If the other side of d(t1) is not in a Reeb annulus there is e
1'0 with d(t) is transverse to
F4S
for t3[t0, t
0#e
1] and [t
0, t
0#e
1]LZ. Notice that the leaf of F4
Sthrough d(t
1) does
not have singularities. Hence the separation property continues to hold for allt(r(s)t
1#e
1which are in Z.
In the case d(t1) is also on the boundary of a second Reeb annulus, then let t
2with d(t
2) in
the other side of the second Reeb annulus. Then Ft1, F
t2are not separated from each other
and Ft1
separates Ft2
from Ft0
— this follows because d(t) is transverse to F4Snear t
1. Hence
Ft1
separates Ftfrom F
t2, for t(t
1, see Fig. 1(b) and the claim holds for all t(r(s)t
2which are in Z.
(ii) Let now s0
be the first time (if any) that d(s0) is a singularity of F4
Sand assume that
the claim holds for all t(r(s)s0
which are in Z. Then d is transverse to F4S
in[s
0!e
2, s
0] and [s
0, s
0#e
2] for some e
2'0. Because d crosses a singularity of F4
Sat time
s0, it follows that oJ (dI ([s
0!e
2, s
0))) is contained in a component of MI !F
s0and
oJ (dI ((s0, s
0#e
2])) is contained in another component of MI !F
s0. As a result the claim holds
for t(r(s)s0#e
2, which are in Z.
Using (i) and (ii) and induction, it follows that the claim holds.
Remark. The restriction to t, r, s3Z is clearly necessary. Suppose that Ft0
and Ft1
are notseparated from each other in the leaf space of FI 4 where t
0(t
1. Then no F3FI 4 separates
834 S. R. Fenley
Fig. 1. (a) Effective representatives of curves in the universal cover; (b) case of adjoining Reeb components.
Ft0
from Ft1, so the claim clearly is false for t
0(t(t
1. This is why (t
0, t
1) is excluded
from Z.
As an immediate consequence of the claim it follows that F1OF
0. Therefore oJ " dI is not
a closed loop in MI , so d is not null homotopic in M.The second possibility for the efficient representative d is that it is contained in a leaf of
F4S. Then d is transverse to F6
Sand the same argument applied to F6
Simplies that d is not
null homotopic.This shows that o(S) is incompressible in M and finishes the proof of the first part of the
theorem.Suppose now that there is a flow line b of '3 intersecting oJ (SI ) in two or more points. Let
p, q3SI with oJ (p), oJ (q)3b. Either p and q are in the same leaf of FI 4Sor there is a path dI LSI
from p to q which is efficient with respect to FI 4S. In the second case the argument above
shows that the endpoints of oJ °dI are in distinct leaves of FI 4, hence not in the same flow lineof '3 . In the first case there is a path from p to q which is transverse to FI 6
S, therefore they
cannot be in the same flow line of '3 either. This shows that every flow line of '3 intersectsoJ (SI ) at most once.
It follows from this that oJ (SI ) is properly embedded in MI and since S is incompressiblethis is (topologically) a plane. This finishes the proof of Theorem 3.3. K
4. PROJECTIONS
In this section we show that # (oJ (SI ))"O if and only if S is a virtual fiber. This is theimportant intermediate step in the proof of the main theorem. First we need the followingeasy lemma.
LEMMA 4.1. ¸et MS
be the cover of M associated to n1(S). ¸et '
Sbe the lift of ' to M
S.
¹hen oJ (SI ) projects to an embedded surface S1
in MS, which is homeomorphic to S and is
transverse to 'S.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 835
Proof. Identify n1(M) with the set of covering translations of n :MI PM. Choosing the
base point p3MI to be in oJ (SI ) but not in the lift of the self intersecting set of o(S), it is easy tosee that up to conjugacy, n
1(S) is exactly the group of covering translations which preserve
the lift oJ (SI ). Since oJ (SI ) is embedded, it now follows that oJ (SI ) projects to an embedded surfaceS1
in MS. As S
1is oJ (SI )/n
1(S), then S
1is homeomorphic to S. By the lifting homotopy
property [19], the map f :SPM lifts to an embedding f1: SPM
Swith image S
1. K
The cover MS
will be used throghout this section. Let nS: MI PM
Sbe the covering
projection and for p3MS, let ¼4
S(p),¼6
S(p) be the lifts of the stable and unstable leaves (of
F4, F6 to MS) which contain p.
THEOREM 4.2. If oJ (SI ) intersects every orbit of '3 then S is a virtual fiber.
Proof. In this proof we do not assume that M is orientable.Every flow line of '3 intersects oJ (SI ) and hence every flow line of ' intersects S. We cut
and paste S (as described by Fried in [16]) preserving the property of being transverse to theflow ' to produce an embedded surface Q transverse to '. It follows that every orbit of' intersects Q.
Given x3Q, if '(0,`=)
(x) intersects Q, let t(x)'0 be the smallest time t with 't(x)3Q.
Otherwise let t(x)"#R. If there is a sequence xi3Q with t(x
i) an unbounded set (including
the case t(xi)"#R for some or all x
i), then take
yi3'
(0,t(xi))(x
i), with '
*~bi, bi+(y
i)WQ"0 and b
iP#R.
Up to subsequence assume that yiPy. The local product structure of the flow implies that
'R (y)WQ"0, contradicting the previous paragraph. It follows easily that Q is a crosssection of '.
Hence, ' is topologically conjugate to a suspension flow of a homeomorphism f ofQ and M fibers over the circle. This f preserves a pair of (possibly) singular foliationsH4, H6 in Q, which are the foliations induced by F4, F6 in Q. In addition some positivepower of f uniformly contracts distances along H4 and uniformly expands distances alongH6. Hence, f is a pseudo-Anosov homeomorphism of a closed surface [37] (or an Anosovhomeomorphism).
There are two cases. First assume that Q is euclidean, in which case f is an Anosovdiffeomorphism of Q. The Klein bottle has only 4 classes of simple closed curves up toisotopy, hence its mapping class group is finite and there are no Anosov diffeomorphisms onthe Klein bottle. This implies that Q is a torus ¹2. By lifting to an orientable double cover ofM if necessary, we may assume that S is orientable, hence a torus. Then o
*(n
1(S)) is a Z=Z
subgroup of n1(M). Since M fibers over S1 with fiber ¹2 and Anosov monodromy, it follows
that the subgroup o*(n
1(S))) has to be a subgroup of n
1(Q). The reason is the following: the
manifold M has a foliation Q by fibers all homeomorphic to Q. Let u :¹2PM be anincompressible immersion with u
*(n
1(¹2))"o
*(n
1(S)) and assume that u(¹2) is in general
position with respect to Q. Eliminate all null homotopic intersections of Q and u(¹2) by cutand past techniques, using that M is irreducible [22]. A first option is that u(¹2) WQ hasa component b which is not null homotopic in Q. In that case u(¹2) can be thought of asa free homotopy from b to itself accross M. It is easy to see that this implies that [b] is equalto tn
*([b]) (in n
1(Q)) for some non zero integer n, where t : QPQ is a monodromy
homeomorphism of the fibration of M with fiber Q [22]. But this is clearly disallowed byt being an Anosov homeomorphism of Q [3]. The remaining possibility is that u(¹2) isdisjoint from Q and contained in Q](0,1). This immediately implies that o
*(n
1(S))Ln
1(Q).
836 S. R. Fenley
Hence, o*(n
1(S)) is a finite index subgroup of n
1(Q). A simple counting argument shows
that there are finitely many subgroups of n1(Q)+Z=Z of a fixed finite index n
0. Therefore,
some power tm of the monodromy satisfies
tm*(o
*(n
1(S))) " o
*(n
1(S)).
This easily implies that there is a finite cover of M to which o*(n
1(S)) lifts to a fiber
subgroup. Consequently S is a virtual fiber. This finishes the proof of the theorem in the caseQ is euclidean.
Now assume that Q is hyperbolic. In that case, the geometrization theorem for fiberingmanifolds [36] implies that M admits a hyperbolic metric. Hence, MI +H3 which iscanonically compactified with a sphere at infinity S2
=. The union H3XS2
=is homeomorphic
to a closed ball BLR3. We will also use the induced Euclidean metric (from R3) in thisclosed ball.
In addition, Cannon and Thurston [6] showed that ' is quasigeodesic (in fact anysuspension is quasigeodesic [40]). Quasigeodesic means that flow lines of '3 are uniformlyefficient up to a bounded multiplicative distortion in measuring distance in (the hyperbolicmetric of) MI . It follows that each flow line c of '3 has well-defined ideal points in S2
=in
forward and backwards time. In addition c is a uniform bounded distance from thecorresponding geodesic of H3 with these ideal points [35]. The bound depends only onM and ' and not on the particular flow line.
Now we show that the limit set of oJ (SI ) in MI XS2=
is S2=. Recall that the limit set of a set
CLH3 is "C"CWS2
=, where the closure is taken in H3XS2
=. Fix a flow line b of '3 which is
the lift of a periodic orbit of '. Let g be the covering translation of MI associated to [n(b)]and b
1be the geodesic of H3 with endpoints the fixed points of g. Let p3S2
=. Since M is
closed its limit set is the whole sphere and it follows that there are covering translates hi(b
1)
of b1
which are arbitrarily close to p in the Euclidean metric of H3XS2=.
Since hi(b) is a uniformly bounded (hyperbolic) distance from h
i(b
1), then h
i(b) is also
arbitrarily near p in the Euclidean metric of H3XS2=. The lift oJ (SI ) intersects every orbit of '3 ,
hence hi(b) intersects oJ (SI ) and oJ (SI ) has points arbitrarily near p (in the Euclidean metric).
Therefore, p3"oJ (SI ) and we conclude that "oJ (SI )"S2=
. By the fundamental dichotomytheorem for surfaces in hyperbolic 3-manifolds (mentioned in the introduction), this impliesthat S is geometrically infinite and therefore a virtual fiber. K
Remarks. The proof of this theorem (for Q hyperbolic) is very unsatisfactory because ita posteriori uses deep geometry results of Thurston and others : the geometrization theoremfor fibering manifolds [36], and the fundamental dichotomy for incompressible immersedsurfaces in hyperbolic 3-manifolds [4, 26, 35]. We believe there must be a proof dependingonly on the topology and the dynamics of the flow, but we do not know how to do that yet.Notice that the fact that the manifold fibers over the circle and the flow is pseudo-Anosov iseasy to prove, reducing it to geometric considerations. Notice also that in the cover M
S,
every flow line of 'S
intersects S1
and S1
is transverse to 'S, therefore every flow line of
'S
intersects S1
exactly once. Hence MS
is homeomorphic to S1]R with a product flow.
This is what is expected when S is a virtual fiber. If for instance one could get a distinctcovering translate (of the group of covering translations M
SPM) of S
1in M
S, it would be
enough to finish the proof. But one does not know that MSPM is a regular cover. In fact
this is equivalent to n1(S) being normal in n
1(M) which would directly imply that S is
a virtual fiber [22].In order to prove the converse of Theorem 4.2, we need to understand what is the set of
orbits of '3 which is intersected by oJ (SI ), or equivalently, the set #(oJ (SI )LO. More specifically
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 837
we need to carefully understand the boundary of this set, that is, L#(oJ (SI ))LO when oJ (SI )does not intersect every orbit of '3 . Let )"#(oJ (SI )). A line leaf of FI 4O is a properlyembedded real line l@ in a leaf l of FI 4O so that the components of l!l@ are all in one side (thatis, a component of O!l@) of l@ in O. The line leaf l@ is regular on a given side » of the setO!l@, if (l!l@)W»"0, that is, there are no prongs of l in ». A line leaf of FI 4 is #~1(l),where l is a line leaf of FI 4O. Similarly define line leaves of FI 6, FI 6O .
The sectors of a leaf l of FI 4O are the closures (in O) of the components of O!l. A leaf isregular if and only if it produces exactly two sectors. Similarly we define sectors for leaves ofFI 4 or FI 6 — these will be three-dimensional closed subsets of MI .
For any set ELO, denote by E]R the set of all y3MI so that #(y)3E.
PROPOSITION 4.3. ¹he boundary of the projection L#(oJ (SI )), is a disjoint union of line leavesof FI 4O, FI 6O , which are regular on the side containing #(oJ (SI )).
Proof. Let p3L) and let pi3) with p
iPp. Let z3MI with #(z)"p. Consider DLMI
a small embedded disk, transverse to '3 with z3int(D) and # injective in D. Let wi3oJ (SI ) with
#(wi)"p
i. By truncating finitely many terms if necessary, there are unique z
i3D and t
i3R
so that wi"'3
ti(z
i). If Dt
iD;#R assume up to subsequence that t
iPt
0. It follows that
wiP'3
t0(z). But since oJ (SI ) is closed in MI , then '3
t0(z)3oJ (SI ), contradicting p N ) .
Assume then that there is a subsequence tiP#R. Suppose that the corresponding
piare all in the same sector » defined by m"#(¼I 4(z)) at p. Let l be the line leaf of m which
bounds this sector.
CLAIM. lLL ) .
Let v3MI with #(v)3l. For i big enough, let qi"¼I uu(v)W¼I 4(w
i). There are
si3R with s
iP#R and '3
si(q
i)3¼I 44(w
i).
Furthermore d('34*(q
i),w
i)P0. This uses the fact that l is regular on the » side, hence
¼I 6(v) intersects ¼I 4(wi) for i big enough. Since oJ (SI ) is transverse to '3 and oJ (SI ) is a lift of
a compact surface, this implies that there are eiP0 with '3
si`ei(qi)3oJ (SI ). Hence for i bigenough #(q
i)3) . In fact this shows that the segment from '3
si(q
i) to w
iin ¼I 44(w
i) projects to
) . Consequently #(v)3) XL) .If #(v)3) , let E be a small disk contained in oJ (SI ) with v in the interior. Hence for i big
enough '3 R(qi)WEO0. There are bounded r
iwith '3
ri(q
i)3E. On the other hand the
argument above shows that there are '3si`ei(qi)3oJ (SI ) with s
i#e
iP#R as iP#R. This
would produce two points
'3ri(q
i) and '3
si`ei(qi) of '3 R(qi) in oJ (SI ),
contradiction. This implies that the stable line leaf lLL) , showing the claim.These arguments also show that l is regular on the side containing ) .A similar argument shows also that given the t
idefined above, if there is some
subsequence tinP!R, then we have an unstable line leaf I through p with ILL ) . Since
) is connected this would force IW)O0 which is impossible. The construction shows thatl is regular on the side containing #(oJ (SI )). This finishes the proof of the proposition. K
Remarks. (1) The fact that tiP#R in the above proof implies the following : if l is
a stable line leaf with lLL ) , then as #(oJ (SI )) approaches l, the points in oJ (SI ) escape in thepositive flow direction. As a consequence it follows that l]R is in the back side of oJ (SI )
838 S. R. Fenley
with respect to the positive flow direction of '3 . The opposite holds if lLL) is an unstableline leaf.
(2) For simplicity we will abuse the notation and say that #(G)LL#(oJ (SI )) when G hasa line leaf G@ with #(G@)LL#(oJ (SI ) ).
We are now ready to show the converse of Theorem 4.2.
THEOREM 4.4. ¹he surface S is a virtual fiber of M over S1 if and only if oJ (SI ) intersectsevery orbit of '3 .
Proof. The if part was proved in Theorem 4.2, so assume that S is a virtual fiber. Theproof will make use of several appropriate covers of M, including the cover M
Sassociated
to n1(S).
We first lift to a finite cover M@ (of M) where S lifts to a surface homotopic to a fiber.Since M@ fibers over the circle, then M@ has a regular cover which is the infinite cyclic coverassociated to this fibering. The infinite cyclic cover has fundamental group n
1(S) and hence
we may assume it is MS.
By Lemma 4.1, the surface S lifts to an embedded compact surface S1"f
1(S) in M
S,
transverse to the lifted flow 'S. Since the group of covering translations of M
SPM@ is
Z and S1
is compact, there is a covering translate S2
of S1
disjoint from S1. The region in
MS
bounded by S1
and S2
is homeomorphic to S1][0,1]. This shows that there is a finite
cover M*
of M so that M*
fibers over the circle with fiber S*
(homeomorphic to S ) andthere is a pseudo-Anosov flow in M
*which is transverse to S
*. The next proposition shows
that S*
is a cross section of the lifted flow in M*. Therefore M
*fibers over the circle with
fiber S*. Notice that up to covering translations of MI we may assume that oJ (SI ) is a lift of
S*
to the universal cover MI . Since S*
is a cross section of the lifted flow, it follows that oJ (SI )intersects every orbit of '3 . This finishes the proof of Theorem 4.4. K
Remark. This theorem shows that if S is transverse to a pseudo-Anosov flow ' and S isa virtual fiber, then ' is topologically conjugate to a suspension flow.
PROPOSITION 4.5. ¸et u be a pseudo-Anosov flow in N so that N fibers over the circle withfiber F and u is transverse to F. ¹hen u is a suspension flow and F is a cross section of u.
Proof. As before, split the singular leaves (if any) of F 4 to produce an essentiallamination V4 in N [18]. This lamination is still transverse to F. Cut N along F to produceF][0, 1]. There is an induced lamination L 4 in F][0, 1], transverse to the boundary andan induced semi flow 'F in L4 which is outgoing along F]M1N and incoming along F]M0N.
Assume that the forward orbit of some point x3(F]M0N)WL 4 does not intersectF]M1N. As seen in the proof of Theorem 4.2, this produces an orbit c of 'F entirelycontained in F](0,1). Cut the original ¼4(c) (which was contained in N) along F and thenlet C be the component containing c. Let v3c.
CLAIM 1. C W (F]M1N)"0.
Suppose not. Let y3CW(F]M1N) and q : [0, 1]PCW(F][0, 1]) with q(0)"v,q(1)"y.Consider
J"Ms3[0,1] D ∀u'0, 'Fu(q(s)) NF]M1NN.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 839
Since F]M1N is transverse to 'F it follows immediatedly that [0,1]!J is open in [0,1].Conversely, since the surface F is transverse to 'F and F is compact then, given s3J there ise'0 so that d('F
t(q(s)),F]M1N)'e for any t'0. Since q(s) and q(s@) are both in the same
stable leaf, the distance between their forward orbits converges exponentially to zero.Hence, if s@ is sufficiently near s then the forward orbits will be (e apart so s@3J and J isopen. Since JO0 (because 03J) and [0,1] is connected; the above facts imply thatJ"[0,1], contradiction to y3F]M1N. This finishes the proof of Claim 1.
Now double F][0,1] along its boundary to produce F]S1. Think of S1 as [!1,1]with the endpoints identified and think of F](!1,1) as embedded in F]S1. LetF0LF]S1 be F]M0N and similarly define F
1. The double of the laminationL4, is denoted
by L4$.
CLAIM 2. ¹he lamination L4$
is an essential lamination in F]S1.
The first case is that there are no complementary components of L4$, that is, L4
$is
a foliation. Then V4 is the foliation F4 and this can only occur when ' is an Anosov flow.Then since F is transverse to an Anosov flow, it has to be a torus or a Klein bottle. Inaddition it is a fiber of a fibration of M over S1. In any case if F is a Euclidean surface, itfollows that n
1(M) is solvable. Under these circunstances Plante [2, 31] proved that ' is
topologically conjugate to a suspension Anosov flow. As seen in the proof of Theorem 4.2, itfollows that F is a cross section of '. This finishes the proof of the proposition in this case.
We can assume from now on thatL 4$is not a foliation and that F is a hyperbolic surface.
Therefore since M fibers over the circle with fiber F, then M is hyperbolic [36]. Notice firstthat the complementary components of L4 are either twisted components or (ideal n-gon)][0,1]. Hence, the complementary components of L4
$are all twisted components (with
n*3). As a consequence, the complementary components of L4$
in F]S1 are irreducible,and their boundaries are incompressible and end incompressible (for detailed definitions ofthese terms see [18]).
There is an induced flow tangent to the leaves ofL4$. Since there are no stationary points
of this flow, it follows that the compact leaves of L4$
have zero Euler characteristic.Therefore there are no sphere leaves in L 4
$.
What remains to be proved is that there are no tori leaves in L4$which bound solid tori.
Assume by contradiction that there is a torus leaf ¹ of L 4$, bounding a solid torus ». Since
M is hyperbolic, the original flow ' is transitive [27] and therefore all the leaves of F 4 aredense in M. Since we are blowing up along singular leaves of F4 to obtainV
s, it follows that
V4 has empty interior and the same is true for L4$. An analysis as in [18] or [30], shows
there is an innermost compressible torus in », that is: there is a torus ¹ @L» (¹ @ may beequal to ¹), so that ¹@ is a leaf of L4
$, ¹@ bounds a solid torus »@, and either
(1) The interior of »@ is disjoint from L4$
or(2) ¹@ has a vanishing cycle [18]. This means that there is a Reeb foliation in the solid
torus »@, so that L4$restricted to »@ is a closed saturated subset of this Reeb foliated
solid torus.
In case (1) the interior of »@ would be a solid torus component of M!L 4$which is not
a twisted component, contradiction. In case (2), since L 4$
has empty interior, there isa complementary component of L4
$contained in »@ which is homeomorphic to R2](0,1),
where R2]M0N and R2]M1N correspond to leaves of L4$
in the interior of »@. Thiscomponent is not a twisted component, again a contradiction.
840 S. R. Fenley
We conclude that there are no tori leaves of L 4$which bound a solid torus in M. These
facts imply that L4$
is an essential lamination in F]S1 [18]. This finishes the proofof Claim 2.
Recall that the leaf C of L4 is contained in F][0,1) and let C2
be the double of C onF]S1. By construction this is a leaf of L 4
$which does not intersect F]M1N. The closure of
C2
is a sublamination C* of L4$which does not intersect F]M1N. Since L 4
$is essential, then
so is the sublamination C* [27]. But F]S1 is a Seifert fibered space, so Brittenham [5]showed that the essential lamination C* contains a sublamination L 4* which is isotopicto either a vertical lamination (that is, a union of fibers w]S1 for some w3F) or isotopic toa horizontal sublamination (that is, transverse to fibers at every point). If L 4* is isotopic toa vertical lamination then clearly L 4*W(F]M1N)O0, contradiction.
Suppose that L4* is isotopic to a horizontal lamination and put it in horizontal formsuch that it is still disjoint from F]M1N. Therefore there is e'0 withL4*LF][!1#e, 1!e]. Let z3L4*, with z in a fiber MwN]S1 so that z satisfies: inMwN]S1, z is a point in L4* which is the closest possible to (w,1). Let ¸
zbe the leaf of
L4* which contains z.Since F is compact, every leaf of a horizontal lamination in F][!1#e, 1!e] will
cover F by the projection map. For any closed loop a in F with basepoint w lift a to a loopa1L¸
zwith starting point z. If a2
1is not closed then either a2
1or a~2
1produces a loop with
final point in w]S1 which is closer to w than z is, contradiction. Hence, the holonomygroup of ¸
zis finite so ¸
zis compact and contained in F](!1, 1) [32]. Since L 4
$is an
essential lamination, ¸zis incompressible [18]. But since F is hyperbolic, n
1(F][!1, 1])
has no Z=Z subgroups. It follows that n1(¸
z) does not inject in n
1(F][!1, 1]) and hence
¸zis not incompressible, contradiction. This finishes the proof of Proposition 4.5. K
Using the techniques of this section, we prove the following proposition, which will beneeded in the completion of the proof of the main theorem later on.
PROPOSITION 4.6. If oJ (SI ) does not intersect every orbit of '3 then there are both stable lineleaves and unstable line leaves in L#(oJ (SI )).
Proof. Suppose that #(oJ (SI ))OO, but L#(oJ (SI )) consists only of stable line leaves. Con-sider the picture in the cover M
Sof M with n
1(M
S)"n
1(S). Recall that M
Shas an embedded
surface S1transverse to the flow '
Sin M
Sand S
1homeomorphic to S. Let B be the subset of
MS
which is the saturation of S1
by the flow 'S.
Since S1
is the projection of oJ (SI ) to MS, the hypothesis that L#(oJ (SI )) has only stable
leaves implies that B is saturated by stable leaves of 'Sand therefore LB is a union of stable
leaves. A boundary component of B is the projection to MSof a component of the boundary
of '3 R(oJ (SI )) in MI . Let ¸ be such a boundary component. Then ¸WS1"0. Let p3¸ and let
pJ 3MI with nS(pJ )"p (where n
Sis the covering projection MI PM
S). Since ¼I 4(pJ ) is a com-
ponent of L'3 R(oJ (SI )), then ¼I 6(pJ ) intersects '3 R(oJ (SI )). Consequently ¼I 6(pJ ) intersects oJ (SI ) andso ¼6
S(p) intersects S
1. If ¸ were on the positive flow side of S
1, then the fact that
¼6S(p)WS
1O0 would imply that the '
Sflow line through p would have to intersect S
1— the
argument is the same as that of Claim 1 in the proof of Proposition 4.5. This is a contradic-tion, which implies that ¸ is on the negative side of S
1.
This shows that the positive flow side N of S1
in MS
is entirely contained in B. Inaddition since S
1separates M
S, it follows that '
Sis a product flow in N, that is, for all u3N
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 841
the flow line through u intersects S1in negative time. In other words N is homeomorphic to
S1][0,#R) and '
Sin N is just the vertical flow.
If there is an upper bound on the distance from points in N to S1, it follows that N is
compact, contradicting 'Sbeing a product flow in N. Consequently, in the universal cover
MI , the component º of MI !oJ (SI ) on the positive flow side of oJ (SI ) contains balls ofarbitrarily large radius. The important conclusion is that the projection of N into M is all ofM. As a result, for each z3M there is t(0 with '
t(z)3S. As in Theorem 4.2, do cut and
paste to S to obtain embedded Q with the same property. We conclude that Q is a crosssection of ' and ' is a suspension flow.
Suppose first that Q is Euclidean. Then ' is topologically conjugate to a suspensionAnosov flow. As seen in the proof of Theorem 4.2, it follows that S is a virtual fiber. But thenTheorem 4.4 implies that SI intersects all orbits of '3 contradiction to hypothesis. Thisfinishes the proof in this case.
The second case is that Q is a hyperbolic surface. It follows that M is hyperbolic [36]and ' is a quasigeodesic flow in M [6]. Since oJ (SI ) does not intersect all of the orbits in MI ,then S is not a virtual fiber by Theorem 4.4. By the fundamental dichotomy for surfaces inhyperbolic 3-manifolds, it follows that S is quasi-Fuchsian.
Let a@ be a regular leaf ofF4Swhich is not in the interior of a Reeb annulus: if there are no
Reeb annuli, a@ can be any regular leaf; if there are Reeb annuli a@ can be a boundarycomponent of a Reeb annulus. Let aLMI be a leaf of FI 4
Swith n(a)"a@ and H3FI 4 with
aLH. The important fact here is that, since L#(oJ (SI )) has only stable line leaves, then#(H)WL#(oJ (SI ))"0. Consequently #(H)L#(oJ (SI )) and as a result #(H)"#(a).
Consider first the situation in SI . Since a@ is not in the interior of a Reeb annulus, itfollows that gJ (a) is a geodesic of SI "H2 with distinct ideal points x, y in the circle at infinityS1=
of LSI "LH2. Choose a parametrization Mps, s3RN of a so that p
sPx if sP#R and
psPy if sP!R. Since S is quasi-Fuchsian, the embedding oJ : SI PMI extends to a continu-
ous embedding u : SI XS1=PMI XS2
=[26, 35]. Then
oJ (ps)Pu(x) if sP#R and oJ (p
s)Pu(y) if sP!R.
In addition since xOy, then u(x)Ou(y).On the other hand if a3a, then a3H and since '3 R(a) is quasigeodesic in H3, there are
z"limt?`=
'3t(a), and z
s"lim
t?~='3
t(p
s).
Notice that all '3 R(ps) are foward asymptotic, having the same forward limit point z3S2
=.
If zsdoes not converge to z as sP#R, then choose s
kwith z
skPvOz. Let b
sbe the
geodesic of H3 with ideal points z and zs. Then b
skconverges to the geodesic b with ideal
points z and v. Since '3 R(psk) are uniform quasigeodesics, they are a uniform bounded
distance from bsk. As b
skconverges to b, then the '3 R(psk
) do not escape a fixed compact setof H3 as kPR. But in H these orbits escape every compact set ! this uses the fact that#(H)"#(a). Hence '3 R(ps
) escapes in H as sP#R, but '3 R(ps) does not escape in MI ,
contradicting the fact that H is properly embedded in MI .We conclude that z
sPz as sP#R and since '3 R(ps
) are uniform quasigeodesics withideal points z and z
s, then '3 R(ps
) converges to z in the Euclidean metric of the unit ballmodel for H3XS2
=. Since oJ (p
s)3'3 R(ps
), it follows that oJ (ps)Pz and therefore z"u(x). In the
same way one shows that z"u(y). This is a contradiction to the previous established factu(x)Ou(y).
We conclude that there must be unstable line leaves in L#(oJ (SI )). This finishes the proofof Proposition 4.6. K
842 S. R. Fenley
Fig. 2. (a) Rectangles; (b) perfect fits in the universal cover.
5. TOPOLOGY OF F4S
AND VIRTUAL FIBERS
This is the most technical section of the paper where we analyse the topology of F4Sand
show that it has closed leaves if and only if S is not a virtual fiber. When F4S
has closedleaves, we show that F4
Sis a finite foliation. First we need some background from the
topological theory of pseudo-Anosov flows [14].If ¸ is a leaf of FI 4 or FI 6 and c is any orbit of '3 contained in ¸, then a leaf piece of
¸ defined by c is a connected component A of ¸!c. The closed leaf piece is AM "AXc andits boundary is LA"c. The flow strip B of ¸ defined by orbits aOb in ¸ is the open intervalof orbits in ¸ with boundary the orbits a and b. The closed flow strip is BM "BXMa,bN.
If F3FI 4 and G3FI 6 then F and G intersect in at most one orbit, since two intersectionswould force a tangency of FI 4 and FI 6. This is easiest seen in O, as FI 4O and FI 6O are thenone-dimensional singular foliations of the plane, having only p prong singularities with p*3.
We say that leaves F,¸3FI 4 and G,H3FI 6 form a rectangle if F intersects both G andH and so does ¸, (see Fig. 2(a)). By Euler characteristic reasons there are no singularities of'3 in the interior of the rectangle region defined by these four leaves. There may besingularities in the boundary sides of the rectangle.
Definition 5.1 (Perfect fits). Two leaves F3FI 4 and G3FI 6, form a perfect fit (F,G) ifFWG"0 and there are leaf pieces F
1of F and G
1of G and also flow strips ¸
1L¸3FI 4 and
H1LH3FI 6, (see Fig. 2(b)) so that, H
1,¸
1,F
1,G
1do not contain singularities of '3 and
M1WGM
1"L¸
1WLG
1, M
1WHM
1"L¸
1WLH
1, HM
1WFM
1"LH
1WLF
1,
∀ S3FI 6, SW¸1O0 NSWF
1O0, (1)
∀ E3FI 4, EWH1O0 NEWG
1O0. (2)
The flow strips ¸1,H
1(or the leaves ¸,H) are not uniquely determined by the perfect fit
(F, G). The implications (1), (2) in fact imply equivalences (that is SW¸1O0 QSWF
1O0
and the same for (2)), see [12, 14]. There is at most one leaf G3FI 6 making a perfect fit witha given leaf piece of F3FI 4 and in a given side of F [11]. Therefore, if (¸,G) forms a perfect fitand g is an orientation preserving covering translation with g(¸)"¸, then g(G)"G.
Definition 5.2. Given p3MI (or p3O) and ¸(p) a leaf piece of ¼I 4(p) defined by '3 R(p), let
J4(¸(p))"MG3FI 6 D GW¸(p)O0N.
Let also
L4(¸(p))"6 Mp3MI Dp3G3J4(¸(p))N.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 843
Then L4(¸(p)) is an open subset of MI and
¼I 6(p)LLL4(¸(p)) so ¼I 6(p) NJ4(¸(p)).
Similarly define J6(¸(p)), L6(¸(p)).
Definition 5.3. Suppose gLF3FI 4 is a (possibly infinite) strong unstable segment sothat for any p3g, there is a leaf piece ¸(p) of ¼I 4(p) defined by '3 R(p), which variescontinuously with p and satisfies
∀ p, q3g, J4(¸(p))"J4(¸(q)).
In that case let P"6p|g¸(p).
Then PLMI is called a stable product region with base segment g. The basis segment isnot uniquely determined by P. Similarly define unstable product regions.
It is easy to see that there cannot be any singularities of '3 in the interior of P. Noticethat for any F3FI 4, G3FI 6 so that (i) FWPO0 and (ii) GWPO0, then FWGO0.
Definition 5.4. (¸ozenges). Let p, q3MI and leaf pieces ¸p, H
pof ¼I 4(p),¼I 6(p) defined by
'3 R(p), leaf pieces ¸q,H
qof ¼I 4(q),¼I 6(q) defined by '3 R(q) so that
J6(¸p)WJ4(H
q)"J6(¸
q)WJ4(H
p).
Then this intersection is called a lozenge A in MI . The set A is an open region in MI . Thecorners of the lozenge are '3 R(p) and '3 R(q), and the sides of A are ¸
p,H
p,¸
q,H
q. The sides
are not contained in the lozenge, but are in the boundary of the lozenge.Sometimes we also refer to p and q as corners of the lozenge.There are no singularities in the lozenges [14]. However, there may be singular orbits on
the sides of the lozenge and the corner orbits also may be singular. The definition ofa lozenge implies that ¸
p, H
qform a perfect fit and so do ¸
q, H
p. This is an equivalent way to
define a lozenge with corners '3 R(p),'3 R(q).Two lozenges are adjacent if they share a corner and there is a stable or unstable leaf
intersecting both of them (see Fig. 3(b)). Therefore, they share a side. A chain of lozenges isa collection MA
iN, i3I, where I is an interval (finite or not) in Z; so that if i, i#13I, then
Aiand A
i`1share a corner (see Fig. 3(b)). Consecutive lozenges may be adjacent or not.
The chain is finite if I is finite.We will also denote by rectangles, perfect fits, lozenges and product regions the
projection of these regions to O. One good way to visualize these objects in O is as follows.Consider proper embeddings m :ºPO of sets ºLR2 into O, sending the horizontal andvertical foliations induced in º to the stable and unstable foliations in m(º)LO. Thena proper embedding is associated to a rectangle m(º) if º"[0,1]][0,1]. A properembedding is associated to a perfect fit if º is a rectangle without a corner, that is,º"[0,1]][0,1]!M(1,1)N. A lozenge is associated to the image of a rectangle without twoopposite corners º"[0,1]][0,1]!M(1,1),(0,0)N (the lozenge is the interior of m(º)).A stable product region is associated to the image of º"[a, b]][0,R) (orº"R][0,R) when the base segment is infinite) and similarly for an unstable productregion. The important fact here is that there are no singular orbits in the interior of any ofthese regions.
844 S. R. Fenley
Fig. 3. (a) A lozenge; (b) A chain of lozenges.
We say that an orbit c of '3 is periodic if it is left invariant by a non-trivial coveringtranslation of MI . The same applies to leaves of FI 4 or FI 6. The main result concerning nonHausdorff behavior of FI 4, FI 6 is the following [14]:
THEOREM 5.5. ¸et ' be a pseudo-Anosov flow in M3. Suppose that FO¸3FI 4 are notseparated from each other. ¹hen F and ¸ are periodic. ¸et »
0be the sector of F containing ¸.
¸et a be the periodic orbit in F and let H0be the component of (¼I 6(a)!a) contained in »
0. ¸et
g be a non-trivial covering translation with g(a)"a and so that g leaves invariant the sectors of¼I 4(a),¼I 6(a). ¹hen g(¸)"¸. ¹his implies that F and ¸ are connected by a finite chain oflozenges MB
iN,1)i)2n, all contained in L6(H
0). Consecutive lozenges are adjacent and all
intersect a common stable leaf E, see Fig. 4 for an example where n"3. All lozenges, sides andcorners are left invariant by g.
Remark. The fact that the MBiN,1)i)2n are adjacent lozenges all intersecting a com-
mon stable leaf implies the following : Let F0"F, F
n"¸ and for 1(i(n, define F
ito be
the stable leaf which is in the boundary of both B2i
and B2i`1
. Then MFiN,0)i)n, are all
non-separated from each other (see Fig. 4). Also for 1)i)n let Gibe the unstable leaf
which is in the boundary of both B2i~1
and B2i
! it is the unique unstable leaf whichseparates F
i~1from F
i. In addition, for each 1)i)n, G
imakes a perfect fit with F
i~1(both
are “asymptotic” leaves in the boundary of the lozenge B2i~1
) and also Gimakes a perfect
fit with Fi(both are in LB
2i) (see Fig. 4). This fact will be essential in the proofs of this
section.
THEOREM 5.6 (Fenley [14]). ¸et ' be a pseudo-Anosov flow in M3 closed. Suppose that' has a product region. ¹hen ' is an Anosov flow (no singularities). In addition ' istopologically conjugate to a suspension Anosov flow.
We now begin to study the topology of F4S.
PROPOSITION 5.7. If S is a virtual fiber then F4S
has no closed leaves.
Proof. Assume that F4Shas a closed leaf l. Let ¼4(l) be the stable leaf containing l. Since
l is transverse to ', it follows that ¼4(l) contains a closed orbit c. Lift to the universal cover
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 845
Fig. 4. The correct picture between non-separated leaves in FI 4.
to get lIL¼I 4(cJ ). Let h3n1(S) be a non-trivial covering translation with h(cJ )"cJ . If
cJ WoJ (SI )O0 choose a point p in this intersection. Since h3n1(S), then h(oJ (SI ))"oJ (SI ). There-
fore h(p) is a different point in the intersection cJ WoJ (SI ) and this contradicts Theorem 3.3. Itfollows that oJ (SI )WcJ"0. Theorem 4.4 then shows that S is not a virtual fiber. K
We will now study what happens when S is not a virtual fiber. Then oJ (SI ) does notintersect every orbit of '3 . In [7] they show that when ' is a suspension pseudo-Anosovflow, then every boundary component of )"#(oJ (SI )) will produce a closed leaf of F4
Sor
F6S. We will obtain the same result here, but we have a much weaker hypothesis. Let
no :oJ (SI )Po(S) be the covering map. This notation will be fixed for the remainder of thissection.
We need a preliminary lemma.
PROPOSITION 5.8. ¸et p be a leaf of FI 4S
with pLZ3FI 4. ¸et G3FI 6 with ZWGO0, sothat there is a line leaf l of #(G) with lLL#(oJ (SI )). Suppose that there is a non-trivial g3n
1(S)
with g(G)"G. ¹hen either no(p) is a closed leaf of F4S, or one of the rays of no(p) limits to
a closed leaf a of F4S
in S. In the second case the ray spirals toward the closed leaf a.
Proof. As G is periodic, let c be the periodic orbit in G. By Proposition 4.3, the leafG"¼I 6(c) is regular on the side containing oJ (SI ), hence ¼I 4(c)WoJ (SI ) is a regular leaf of FI 4
S.
Then
g(¼I 4(c))"¼I 4(c), g(oJ (SI ))"oJ (SI ) N g(¼I 4(c)WoJ (SI ))"¼I 4(c) WoJ (SI ).
Hence a"n(¼I 4(c)WoJ (SI )) is a closed leaf of F4S. Assume that g acts as a contraction in
the set of orbits in ¼I 6(c). Fix p3¼I 4(c)WoJ (SI ). We refer to Fig. 5.If Z"¼I 4(c), then p"ZWoJ (SI )"¼I 4(c)WoJ (SI ) and therefore no(p)"a is a closed leaf of
F4S. Suppose then that Z is not equal to ¼I 4(c). Then ZWGO0, implies that
gn(Z) P ¼I 4(c) as nP#R
in the leaf space of FI 4. There may be other leaves in the limit as well.This convergence property and the fact that g(oJ (SI ))"oJ (SI ) imply that for n big enough
then gn(p)LoJ (SI ) passes through w3¼I 6(p)WoJ (SI ), which is arbitrarily close to p. Let q be theray of p defined by g~n(w) and so that #(q) limits on #(G). The rays q and gn(q) project to thesame leaf of F4
S. Let b be the segment of ¼I 4(c)WoJ (SI ) between ¼I 6(p) and ¼I 6(g~1(p)). Then
a"no(b).Since g~n(¼I 6(p)) converges to G"¼I 6(c) as nP#R, then q is the union 6
m*0H
m,
where Hm
is the piece of q between g~n~m(¼I 6(p)) and g~n~m~1(¼I 6(p)). Let Em
be the piece of
846 S. R. Fenley
Fig. 5. Producing the spiralling effect in the leaves.
gn`m(p) between ¼I 6(p) and ¼I 6(g~1(p)). Then Em"gn`m(H
m). It follows that
no(q)"6m*0
no(Hm)"6
m*0no(Em
)
because no(Em)"no(Hm
). But Em
converges to b as mP#R. We conclude that no(q) spiralstowards a. This finishes the proof of the proposition. K
The first step in the proof of the main theorem is to analyse the possibilities for thetopology of the geodesic lamination G4 in S associated to F4
S. Similarly for the unstable
foliation and lamination. The key result we will need is the following
THEOREM 5.9. Either every leaf of G4 is dense in G4 or every minimal set of G4 is a closedleaf of G4. In other words, either G4 is minimal or F4
Sis a finite foliation.
Proof. Given the structure of foliations and geodesic laminations in the torus or Kleinbottle [21], this result is very easy if S is Euclidean, therefore assume from now on that S ishyperbolic. The proof will consist of understanding how FI 4
Ssits in the universal cover, with
relation to FI 4.Assume therefore that there is a minimal sublamination G4*, which is not a closed
geodesic. The goal is to show that G4*"G4. LetF*"g~1(G4*) (recall that g is the map which
pulls tight the leaves of F4S
or F6S
which are not in Reeb annuli). Then F*LF4
Sand
F*
does not have a a closed leaf. Notice that by definition every leaf of G4* is dense in G4*.However the same does not follow for F
*, because there may be thick leaves (of G4) in G4*.
This will be the main difficulty in the proof of the theorem.Two facts from Section 2 will be used throughout the proof : first there is at most one
singularity in a leaf of F4S; second a non-simply connected leaf of F4
Sdoes not contain
a singularity and is a closed loop. Consequently, the geodesic laminationG4 has only simplyconnected (ideal polygon) complementary components in S.
For e'0 sufficiently small the e neighborhood of G4* in S is an embedded subsurfaceB of S, whose topological type is independent of e. Let f be a component of LB. Then f is an
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 847
Fig. 6. (a) Asymptotic leaves in the lamination; (b) Bands of doubly asymptotic leaves in the foliation.
embedded curve in S which follows finitely many boundary leaves c1,2,c
kof G4*. This is
because the component of S!G4* containing f has finite hyperbolic area. Notice thatconsecutive c
i@s are asymptotic in S.
The strategy is to show that f is null homotopic in S. This will imply that the onlypossible leaves of G4!G4* are in the interior of ideal polygon complementary regions of G4*.In the foliation setting they would correspond to leaves of F4
S“crossing a singularity”. But
in the construction of G4, the leaves of G4 never come from leaves of F4S
which crossa singularity, hence G4"G4*. To show that f is null homotopic we will extensively workback and forth between the foliation and lamination settings.
Since there are finitely many complementary regions of G4 all of which have finitehyperbolic area, it follows that there are at most finitely many leaves of G4 which areasymptotic to c
1and c
2and in between c
1and c
2. This is because each additional leaf
produces a cusp in the hyperbolic surface S!G4, and in a hyperbolic surface of finite areathere can only be finitely many cusps. Let a
0"c
1,a
1,2,a
k"c
2be these leaves, with
ailocally separating a
i~1from a
i`1(see Fig. 6(a)). The idea is to show that all a
ihave
identified rays in the foliation setting. Going around the components ci will show that f ishull homotopic.
Lift to oJ (SI ) to obtain asymptotic aJ0,2,aJ
k. Let ½
i,0)i)k be the bands of FI 4
Sin SI with
gJ (v)"aJifor any v3½
i. Here bands is used in a generalized sense : if a
iis not thick then ½
iis
a single leaf. Notice that the “last” leaf v1
of ½iand the “first” leaf v
2of ½
i`1share a ray and
are contained in a singular leaf m of FI 4S(see Fig. 6(b)). In generating GI 4, the leaf m is split to
produce aJiand aJ
i`1(and other leaves too). This works for 0)i(k.
The main technical result is the following:
PROPOSITION 5.10. None of the leaves ai, 0)i)k is a thick leaf of G4.
Proof. The proof will be by contradiction. Suppose that some ajis a thick leaf of G4.
Parametrize the leaves of FI 4S
in
½" 60)i)n
½i
as MRtD 0)t)1N
with R0
the leaf containing the “first” leaf of ½0
and R1
containing the “last” leaf of ½k.
Notice that the “last” leaf of ½iand the “first” leaf of ½
i`1are subsets of a single leaf of
FI 4Sbecause they have a common ray. The set ½ is a non degenerate interval of leaves by the
848 S. R. Fenley
Fig. 7. Forcing boundary leaves to be periodic.
assumption that there is a thick leaf in Z0)i)k
ai. Each ½
iprojects homeomorphically to
S because all leaves in no(½i) are asymptotic to a minimal non-closed leaf sublamination.
Therefore, as one moves out the ends ½i, the thickness decreases to zero. This implies that in
the ends each unstable leaf of FI 6S
intersecting R0
will intersect all Rt. Let q
0be such an
unstable segment intersecting all the bands ½i. Assume also there are no singularities in
these deep subsets of ½i(this can be done because there is at most one singularity in a leaf of
FI 4S). Define p
t"q
0WR
t. Let MI
0be the component of MI !¼I 6(p
0) containing the subray of
Rt. Let ¸
tbe the leaf piece of ¼I 4(p
t) defined by p
tand contained in MI
0. Then ¸
tcontains the
ray of Rtdeep in the ends of ½
i(here we use that p
tis not a singularity). Let
pt" ¸
tWoJ (SI ) LR
tWoJ (SI ) a ray of FI 4
S.
Then no(pt) is a ray of F4
S. Notice that Z
0)t)1ptis a “product” region of FI 4
Sin oJ (SI ). In
addition all ptare asymptotic rays of FI 4
S.
CLAIM 1. #(¸0)WL#(oJ (SI ) )"0.
Suppose not. Let G3FI 6 with #(G)LL#(oJ (SI )) and ¸0WGO0. The lamination G4* is
minimal hence the leaf no(a0) limits on itself on the side opposite to the other leaves no(ai),i*1. Therefore no(½0
) limits on itself on the side opposite to no(½i). Since ½
0may be a thick
leaf we need to understand what this means. The thickness of ½0
decreases to 0 moving outthe ends. Hence, no(½0
) will limit on one of the boundary leaves of no(½0). It is clear that it
can only limit on the outer leaf, which is exactly no(p0). It follows that no(p0
) limits on itself.Choose w3p
0arbitrarily far from p
0in p
0and h
13n
1(S) with h
1(w) arbitrarily near p
0. Since
¼I 4(p0)WGO0 then for h
1(w) sufficiently near p
0, it follows that ¼I 4(h
1(w))WGO0 (see
Fig. 7). But h1(oJ (SI ))"oJ (SI ), so h
1(L#(oJ (SI )))"L#(oJ (SI )). Hence, the boundary component of
#(oJ (SI )) intersecting #(¼I 4(p0)) is taken by h
1to the boundary component of #(oJ (SI ))
intersected by #(¼I 4(h1(w))). In both cases this boundary component is G, which implies that
h1(G)"G. Proposition 5.8. then shows that no(p0
) is either closed or spirals towardsa closed leaf of F4
S, so in any case it has a closed leaf in its closure. Since no(p0
)Lg~1(G4*),the above fact implies that G4* has a compact leaf which is a closed geodesic. Thiscontradicts the fact that G4* is a minimal lamination which is not a closed geodesic leaf.Therefore #(¸
0)WL#(oJ (SI ))"0, proving Claim 1.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 849
In particular, this implies that p0"¸
0WoJ (SI ) and that #(¸
0)L#(oJ (SI )). The same is true
for ¸1, so p
1"¸
1WoJ (SI ) and #(¸
1)L#(oJ (SI )). In addition this implies that J4(¸
0)"J4(¸
1).
Notice that in this claim it was not assumed that L#(oJ (SI ))O0.
CLAIM 2. ¹here is 0(t(1 so that #(¸t)WL#(oJ (SI ))O0.
Suppose not. Then for any 0)t)1, #(¸t)WL#(oJ (SI ))"0. This is equivalent to
#(¸t)"#(p
t) for any t. Therefore ¸
tdoes not contain a singular orbit of '3 , because there
are no singularities in pt. In addition J4(¸
t)"J4(¸
t{) for any 0(t, t@(1, because there is
a product structure of FI 4S
in X0)t)1
ptand #(¸
t)"#(p
t). As a result M¸
tD 0)t)1N
forms a stable product regionP in MI with '3 R(q0W½) as a base flow strip. This is disallowedby Theorem 5.6. This proves Claim 2.
Notice that Claim 2 implies that L#(oJ (SI ))O0. This is a consequence of the assumptionthat some a
iis a thick leaf of G4.
From now on fix parametrizations of ptas Mp
t(s) D s3[0,#R)N, so that for any s*0 and
any 0)t, t@)1, pt(s) and p
t{(s) are in the same unstable leaf of FI 6
S.
By Claim 2 there is G3FI 6 with #(G)LL#(oJ (SI )) and G intersecting some ¸t1. This
implies that ¼I 6(pt1(s))PG in FI 6 as sP#R.
This leaf G with #(G)LL#(oJ (SI )), will be the cornerstone of all the constructivearguments in the rest of the proof.
The same argument as in the proof of Claim 1, produces w3p0
far away from p0
andh3n
1(S) a covering translation of oJ (SI ) with h(w)3¼I 6(p
0)WoJ (SI ) and very near p
0. A priori
this is not a contradiction, since ¼I 4(p0) does not intersect any G3FI 6 with #(G)LL#(oJ (SI )).
Let Z"h(¸0)WMI
0(recall that MI
0is the component of MI !¼I 6(p
0) containing p
0). Then
Z does not contain a singularity, because ¸0
does not. There are 4 cases to consider, each ofwhich will lead to a contradiction.
Case 1 : J4(¸0)(J4(Z). Recall that J4(¸
0)"M¼I 6(p
0(s)) D 0(s(#RN. In addition
¼I 6(p0(s))"¼I 6(p
t1(s))PG as sP#R. Since J4(¸
0)(J4(Z), there is H3FI 6 with
HLLL4(¸0) and H3J4(Z). Therefore ¼I 6(p
0(s))PH as sP#R and it follows that H is
not separated from G in the leaf space ofFI 6. By Theorem 5.5, H is connected to G by a finitechain of adjacent lozenges MB
iN,1)i)2n, all intersecting a common unstable leaf. As
described in the remark after Theorem 5.5, there are unstable leaves G"G0,2,G
n"H, in
the boundary of these lozenges, with MGiN,0)i)n, non-separated from each other; see Fig.
8 for an example with n"4. Also there are stable leaves MFiN,1)i)n, with F
ithe unique
stable leaf separating Gi~1
from Gi. Then F
imakes a perfect fit with G
i~1and also with G
i.
Suppose that ¸0intersects one of the lozenges in the chain from G to H. By construction
¸0WG"0 and ¸
0WH"0, therefore there is j with 1)j(n so that ¸
0intersects either
B2j
or B2j`1
. Therefore ¸0WG
jO0. By Claim 1 above p
0"¸
0WoJ (SI ), so p
0WG
jO0. Let
s* with p0(s*)3G
j. In that case G
jwill not intersect ¼I 4(p
t1(s*)), because ¼I 4(p
t1(s*))WG
0O0
and G0,G
jare not separated in the leaf space of FI 6. But this contradicts the fact that by
definition ¼I 6(pt1(s*))"¼I 6(p
0(s*))"G
jand ¼I 6(p
t1(s*)) obviously intersects ¼I 4(p
t1(s*)).
Therefore,¸0does not intersect the interior of the lozenges, but the stable leaf containing ¸
0separates G from H. As a result ¸
0has to be contained in one of the leaves MF
iN,1)i)n,
say ¸0LF
i0. In the example of Fig. 8 (with n"4), then ¸
0LF
2so i
0"2.
For 1)i)n, let ki"(F
iWMI
0WoJ (SI )), so p
0"k
i0.
The construction implies that #(G)LL#(oJ (SI )) and HWoJ (SI )O0, because#(Z)L#(oJ (SI )) and HWZO0. Since oJ (SI ) is connected and all F
iseparate G from H, it
850 S. R. Fenley
Fig. 8. The case J4 (¸0)(J4 (Z ).
follows that FiWoJ (SI )O0 for any 1)i)n. Since H intersects oJ (SI ), then there is a leaf piece
H@ of H (without singularities) making a perfect fit with Fn
and so that #(H@)L#(oJ (SI )).Otherwise there is x3H, with #(x)3L#(oJ (SI )). But since H is an unstable leaf and inter-sects oJ (SI ), it follows from Proposition 4.3 that there is a line leaf X
0of ¼I 4(x) with
#(X0)LL#(oJ (SI )). But then X
0would separate HWoJ (SI ) from F
i, contradiction to F
iinter-
secting oJ (SI ). This proves the existence of such a leaf piece H@ making a perfect fit with Fn. In
addition, if #(FnWMI
0) intersects L#(oJ (SI )), then a similar reasoning produces an unstable
line leaf X1
with X1
separating oJ (SI ) from H, contradiction.We conclude that the perfect fit (F
n,G
n) projects to oJ (SI ), producing two leaves
kn"F
nWoJ (SI ) of FI 4
Sand (G
nWoJ (SI )) of FI 6
Swhich form a “perfect fit” in oJ (SI ). Perfect fits of
leaves of FI 6S, FI 4
Sin oJ (SI ) are defined in the same way as in MI . We refer to Fig. 13(a) with
kn"l and G
nWoJ (SI )"l@.
Perfect fits of leaves of FI 4S, FI 6
Sin oJ (SI ) are very rare and only occur in very special
circumstances. In Lemmas 5.13 and 5.14 we show that the rays in the perfect fit project toS as either closed leaves or rays spiralling to closed leaves of respective foliations. For thesake of continuity, Lemmas 5.13 and 5.14 are located at the end of this section.
Therefore no(kn) is either closed or spirals towards a closed leaf of F4
S. In either case let
no(k) be this closed leaf, where either kn"k or k
nis asymptotic to k (in the case no(kn
) spiralstowards no(k)). Let g3n
1(S) be the covering translation of oJ (SI ) associated to no(k) with
a fixed point c@3S1=
so that knLoJ (SI ) has an ideal point c@3S1
=. Assume that c@ is the
repelling fixed point of g.Consider the leaves MG
iN,i
0)i(n. First suppose that none of them intersects oJ (SI ). This
implies that the region in oJ (SI ) bounded by
Fi0WoJ (SI ), ¼I 6(p
0)WoJ (SI ) and F
nWoJ (SI )
is free of singularities of FI 4S
and FI 6S— this region is contained in the union of the lozenges
Bi, 2i
0)i)2n!1 and its unstable boundary sides. The boundary rays of this region are
ki0
and kn. As in the proof of Lemma 5.14, the singularity-free property implies that the
boundary rays ki0
and knhave the same ideal point in S1
=, which is c@. If no(ki0
) is contained inthe interior of a Reeb component, then it spirals towards a closed leaf of F4
S. Suppose this is
not the case. Then R0, the leaf of FI 4
Scontaining k
i0is a K-quasigeodesic for some K and so
are gm(Rt) for any m3Z. When mP#R, one of the ideal points of gm(R
0) is c@ and the other
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 851
Fig. 9. The case J4(¸0)'J4(Z).
converges to the attracting fixed point of g. Since gm(Rt) are uniform K-quasigeodesics for
any m3Z, this implies that gm(ki0) converges to a leaf k* of FI 4
Swhen mP#R. In addition
g(k*)"k*, so no(k*) is a closed leaf of F4S. Since gm(k
i0)Pk* when mP#R, then no(ki0
)limits in no(k*). This implies that in fact no(ki0
) spirals towards the closed leaf no(k*). So inany case no(p0
)"no(ki0) is either closed or spirals towards a closed leaf of F4
S. As seen in the
proof of claim 1, this is a contradiction to G4* not containing closed geodesics.The second possible option is that there is i
1with i
0)i
1(n so that G
i1intersects oJ (SI ).
Assume i1
is the smallest possible. As in the case for Gn"H (with G
nWoJ (SI )O0), this
implies that no(ki1) is either a closed leaf of F4
Sor spirals towards a closed leaf ofF4
S. We can
now apply the argument in the previous paragraph with ki1
instead of kn
to arrive ata contradiction.
This shows that case 1 cannot occur.Case 2: J4(¸
0))'J4(Z). In a similar way to case 1, there is ¹3FI 6, with ¹LLL4(Z)
and ¹3J4(¸). Then the leaf ¹ is not separated from h(G) and ¹W¸0O0, (see Fig. 9). Recall
that J4(¸0)"J4(¸
1) (Claim 1), therefore ¹W¸
1O0. We can now apply the proof of
case 1 with h(¸1)WMI
0in the place of ¸
0, ¸
1in the place of Z, h(G) in the place of G and ¹ in
the place of H. The arguments in case 1 then show that no(h(p1))"no(p1
) is either closedor spirals towards a closed leaf of F4
S. As in case 1 this is a contradiction. Therefore
case 2 cannot happen either.Case 3: J4(¸
0)"J4(Z). First recall that ¼I 6(p
0(s))PG when sP#R. Therefore
»s"h(¼I 6(p
0(s)))Ph(G) when sP#R. But
»s" h(¼I 6(p
0(s))) " ¼I 6(h(p
0(s)))
and h(p0(s)) escapes in ZLh(¸
0) as sP#R. Since J4(¸
0)"J4(Z) it follows that
¼I 6(h(p0(s))) escapes J4(¸
0) as sP#R, hence »
s"¼I 6(h(p
0(s))) converges to G also.
Therefore »sconverges to both G and h(G) when sP#R and we conclude that G and h(G)
are not separated in the leaf space of FI 6.Let G"G
0,2,G
n"h(G)"H be the sequence of leaves of FI 6 which are not separated
from G and are in the unstable boundaries of the lozenges in the chain from G to h(G). For1(i(n, let F
i3FI 4 be the only stable leaf separating G
i~1from G
i; see Fig. 10 for
an example with n"4. Let also MBi,N, 1)i)2n be the chain of lozenges from the
periodic orbit l0
in G to the periodic orbit l1
in H — they all intersect a common unstableleaf ¼I 6(p
0).
852 S. R. Fenley
Fig. 10. The case J4(¸0)"J4(Z) produces non-separated leaves in the universal cover.
As seen in the proof of case 1, ¼I 4(p0) is one of F
iso we may assume that G is chosen so
that ¼I 4(p0)"F
1and therefore ¸
0makes a perfect fit with G. Then ZLh(¸
0) makes
a perfect fit with h(G) (see Fig. 10).Suppose there is 1(i
2(n with G
i2WoJ (SI )O0 and assume i
2is the smallest such
number. Then Gi2
forms a perfect fit with Fi2
and this produces a perfect fit in oJ (SI ). A proofas in case 1 shows that this is not allowed. Consequently all G
isatisfy #(G
i)LL#(oJ (SI )).
The leaves h(¸0) and H make a perfect fit and are both in the boundary of a lozenge
B2n`1
"h(B1), which is adjacent to B
2n. Then ¼I 6(p
0) intersects the interior of this lozenge,
so MBi, 1)i)2n#1N, forms a chain of adjacent lozenges, all intersecting a common
unstable leaf ¼I 6(p0). Notice that ZLh(¸
0)L¼I 4(h(p
0)).
The region of MI bounded by ¼I 4(p0)"F
1, ¼I 6(p
0), ¼I 4(h(p
0)) and G
1,2,G
nis void of
singularities; see Fig. 10 — this region is contained in the union of lozenges above and theirsides. This projects to a singularity free region of oJ (SI ). The fact G
iWoJ (SI )"0 for any i,
then implies as in the proof of Lemma 5.14, that the boundary rays of this region have thesame ideal point in S1
=. These boundary rays are k@"(F
1WMI
0WoJ (SI )) and
h(k@)WMI0"(h(F
1)WMI
0WoJ (SI )) and they have the same ideal point c3S1
=. As a consequence
h(c)"c. If no(k@) is contained in a Reeb annulus then it spirals towards a closed leaf of F4S.
Otherwise k@ is a quasigeodesic in oJ (SI ) and an argument as contained in case 1 shows thefollowing : either no(k@) is a closed leaf or no(k@) spirals towards a closed leaf of F4
S. As in
case 1 this leads to a contradiction. This shows that case 3 cannot happen either.Case 4: J4(¸
o) and J4(Z) are not comparable. This is the last case. If no(p0
) entersa Reeb annulus C@ of F6
S, then no(p0
) cannot exit C@ and has to limit in a closed leaf of F4S,
contradiction as seen before. The same occurs if no(p0) is contained in a Reeb annulus.
Therefore, by taking a subray if necessary, we may assume that no(p0) does not intersect
a Reeb annulus of F6S.
Since J4(¸0) and J4(Z) are not comparable, there are A, B3FI 6 so that ALLL4(Z)
with A3J4(¸0) and BLLL4(¸
0) with B3J4(Z) (see Fig. 11). It follows that A and B are not
separated from each other in the leaf space of FI 6.Therefore since AW¸
0O0 and BWZO0, then A and B intersect oJ (SI ) and A
1"AWoJ (SI ),
B1"BWoJ (SI ) are distinct leaves of FI 6
S. In addition since #(oJ (SI )) contains #(¸
0) and #(oJ (SI ))
contains #(Z) (see remarks at the end of claim 1 above), it follows that A1
and B1
are notseparated in FI 6
S. As explained in Section 2 this can only happen when no(A1
) and no(B1) are
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 853
Fig. 11. The non-comparable case produces non-separated leaves of FI 4.
the boundaries of a Reeb annulus C in F6S. Let CI be the lift of C to oJ (SI ) with boundary
A1XB
1.
The construction implies that ¼I 6(p0)WoJ (SI ) is contained in the region of oJ (SI ) bounded
by A1"AWoJ (SI ) and B
1"BWoJ (SI ). This region is CI . Therefore
no(¼I 6(p0)WoJ (SI ))LC so no(p0) 3 no(¼I 6(p0)WoJ (SI ))LC.
As a result no(p0) is contained in a Reeb annulus. But no(p0)3no(p) and this contradicts thefact that by assumption no(p0
) does not intersect a Reeb annulus.This contradiction shows that case 4 cannot happen either.Since none of the cases 1—4 can happen, we conclude that the assumption that some a
iis thick is impossible. This finishes the proof of Proposition 5.10. K
Conclusion of the Proof of ¹heorem 5.9. Proposition 5.10 shows that aiis not a thick
geodesic of G4 for any 0)i)k.This implies that in the foliation setting the leaves, g~1(a
1),2,g~1(a
k~1) are single
leaves of F4S
and so have rays which are all consecutively identified to each other. Sincethere is at most one singularity of FI 4
Sfor each stable leaf, it follows that in F4
S, the
singularity in g~1(a0) is identified with the singularity in g~1(a
1). By induction the singu-
larity in g~1(a0) is identified with the singularity in g~1(a
k).
We now analyze the boundary leaves ciof G4* associated to the boundary component
f of the e neighborhood of G4* (see Fig. 12). Then g~1(a0)"g~1(c
1) and g~1(a
k)"g~1(c
2),
so no(R0)Lg~1(c
1) and no(R1
)Lg~1(c2). With all the arguments above we have shown that
the singularity x1
(in the foliation setting) coming from g~1(c1) is identified with the
singularity x2
coming from g~1(c2). Going around the boundary components c
2,c3,2
associated to f, induction shows that all singularities xiare identified to a unique singular
point x in the foliation setting (see Fig. 12). We conclude that the loop f is null homotopicin S. Therefore, the complementary regions of G4
*are simply connected and ideal polygons.
This also shows that there are no other leaves in G4 (they would have to cross singularitiesin F4
S). Hence, G4
*"G4 and G4 is a minimal lamination.
We conclude that if G4 has a minimal sublamination which is not a closed leaf, then G4 isminimal. This finishes the proof of Theorem 5.9. K
854 S. R. Fenley
Fig. 12. Collapsing boundaries of complementary components of sublamination.
Given this result we can completely understand the case #(oJ (SI ))OO and finish the proofof the main theorem.
THEOREM 5.11. If S is not a virtual fiber then the only minimal sets of F4Sare closed leaves
of F4S. Equivalently every ray of a leaf of F4
Sspirals towards a closed leaf of F4
S. In addition,
there are finitely many closed leaves in F4S. Similarly for F6
S.
Proof. Since S is not a virtual fiber, Theorem 4.4 shows that #(oJ (SI ))O0. Proposi-tion 4.6 in fact shows that there stable and unstable line leaves in L#(oJ (SI )). Given G3FI 6with a line leaf lL#(G) satisfying lLL#(oJ (SI )), we will then show that F4
Sis a finite
foliation. Using D*3FI 4 with a line leaf m*L#(D*)WL#(oJ (SI )), the same proof shows thatF6
Sis also a finite foliation.The proof will be by contradiction. Suppose then that F 4
Sis not a finite foliation. By
Theorem 5.9. it follows that G4 is a minimal lamination. Let z3oJ (SI ) with ¼I 4(z)WGO0.Let E be the leaf of FI 4
Scontaining z. If gJ (E) is not a thick geodesic, then as in the proof of
Claim 1 of Proposition 5.10 (which works exactly the same if S is euclidean) we arrive ata contradiction to the fact that G4 is a minimal lamination which is not a closed leaf. If, onthe other hand, gJ (E) is thick, consider the band ½ of FI 4
Scontaining E. As seen in the proof of
Claim 1 of Proposition 5.10, there is a boundary leaf F on one side of the band ½, so thatno(¼I 4(z)WoJ (SI )) limits on no(F) in S. The arguments of Cases 1—4 of Proposition 5.10 (whichalso work if S is euclidean) produce a contradiction as above. This contradiction shows thatin fact F4
Shas to be a finite foliation.
Suppose now that there are infinitely many closed leaves. In any case there are onlyfinitely many isotopy classes of closed leaves. Let p be a limit point of an infinite set Mb
iN,
i3N of isotopic closed leaves. Let b be the leaf through p. Then b is also closed. This followsfrom an argument of Haefliger, [20] used for codimension one foliations: limits of compactleaves are compact. Exactly the same proof works for singular foliations.
Then ¼4(b) has fundamental group Z. Given the structure of the foliation F4 aroundperiodic orbits ' and non-simply connected leaves of F4, we can choose the orientation ofb so that F4 has contracting holonomy associated to b. This implies that there are no closedleaves of F4
Ssufficiently near b, which is a contradiction to the assumption. This finishes the
proof of the theorem. K
Finally we have the following:
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 855
Fig. 13. (a) A perfect fit in SI , (b). rays with distinct ideal points produce big regions in SI without singularities.
THEOREM 5.12. If S is a virtual fiber, then every leaf of F 4Sis dense in S and consequently
F4S
does not have closed leaves.
Proof. Since S is a virtual fiber, the proof of Theorem 4.4 shows that: (1) S lifts to anactual fiber in a finite cover of M and (2) this fiber is a cross section of the lifted flow in thefinite cover of M. In addition the lifted flow is a suspension pseudo-Anosov flow. If leaves ofF4
Sare dense in a finite cover, then clearly they are also dense in SLM, so we may assume
that S is an actual fiber and ' is a suspension pseudo-Anosov flow. Given that, the situationis well understood by the classical study Anosov diffeomorphisms of the torus andpseudo-Anosov diffeomorphisms of closed hyperbolic surfaces [3]. In particular, there areno doubly asymptotic leaves in FI 4
S[3], hence there are no bands of FI 4
Sand no thick leaves
of G4. Also every leaf of G 4 is dense in G4 [3] and since there are no thick leaves in G4, itfollows that every leaf of F4
Sis dense in S. This finishes the proof of the theorem. K
We now prove that perfect fits between leaves of FI 4S, FI 6
Sonly occur in special
circumstances, which was needed for the proof of Proposition 5.10.
LEMMA 5.13. Suppose that S is a hyperbolic surface. If l and l@ are rays in leaves of FI 4S,
FI 6S
respectively, so that l and l@ form a perfect fit, then either l and l@ project to closed leaves ofF4
S, F6
S, or they project to leaves spiralling to closed leaves of F4
S, F6
S,
Proof. Let ¼I 6S(x) be the leaf ofFI 6
Sthrough x and similarly define ¼I 4
S(x). Let a"no(l). If
a is closed we are done. Therefore assume a is not closed.Since l and l@ form a perfect fit, there are segments z
1,y
1in leaves ofFI 6
S, FI 4
S, respectively,
so that the properly embedded curve
f@ " l X z1
X y1
Xl@
“encloses” the perfect fit (see Fig. 13(a)). The segments z1,y
1take the role of the flow bands in
the definition of perfect fits in MI . The important fact is that the perfect fit region ofSI bounded by f@ has no singularities of FI 4
S,FI 6
S.
Let w0
be the starting point of l@. Let ui3l escaping in l and let w
i"¼I 6
S(u
i)W¼I 4
S(w
0).
Then wiPw
0as iP#R. Let [u
i,w
i]ube the segment in ¼I 6
S(u
i) from u
ito w
i. Suppose there
is a subsequence MikN, k3N where the lengths of these segments are bounded. Since w
iPw
0,
then these segments would converge in SI . Then uikwould converge in SI contradiction to the
fact that l escapes in SI .
856 S. R. Fenley
We claim the following: Let b be a leaf in the limit of a and bI a lift to SI . Then in at leastone side of bI , all rays of unstable leaves starting in bI do not have singularities. The reason isthe following: the ray a is limiting on b, so there is a local side of b, where bigger and biggerunstable segments starting on that side do not have singularities. This is because theunstable segments [u
i,w
i]udo not have singularities and their lengths converge to #R. If
there is a point v3b and a ray of ¼I 6S(v) on that side with a singularity, we obtain
a contradiction. This proves the claim.The following lemma proves that either b is closed or spirals towards a closed leaf ofF 4
S.
Since a is limiting to b it easily follows that a has to limit in a closed leaf of F4S. Because S is
two-dimensional this implies that in fact a spirals towards a closed leaf of F 4S. This finishes
the proof of Lemma 5.13. K
LEMMA 5.14. Suppose that S is a hyperbolic surface. ¸et c be a leaf of FI 4Swith a given side
so that every ray of FI 6S
starting in c and in that side does not contain a singularity. ¹henc projects to either a closed leaf of F6
Sor a leaf with a ray spiraling towards a closed leaf.
Proof. Notice that a ray d of FI 6S
has a well defined limit point at infinity (S1=). This is
true even if no(d) is contained in the interior of a Reeb annulus because then no(d) spiralstowards a closed leaf.
We claim that all rays of FI 6S
as above starting in c have the same ideal point in S1=.
Suppose not, say q1,q
23c have rays r(q
1),r(q
2) with distinct ideal points c
1,c2(see Fig. 13(b)).
Consider the region D of SI +H2, which is bounded by
q"r(q1)X r(q
2) X [q
1,q
2]s
with D the component of SI !q not containing rays of c.Then D does not contain any singularity : the rays r(q) for q between q
1and q
2, do not
contain any singularity in their interior. Hence the union D2"Z
q|(q1,q2)sr(q) does not
contain any singularity. Any leaf in the boundary of D!D2
would have a singularity andbe contained in r(q) for some q. We conclude that D does not have any singularity.
Let now c@3S1=
between c1
and c2
and in the closure of D. Then there is a half planeD
*LH2 centered at c@ and contained in D. Since there are disks of arbitrarily large radius
contained in D*
it follows by projection to S, that there are no singularities in F6S. This is
a contradiction and proves the claim. Therefore all rays r(q),q3c have the same ideal pointc3S1
=. This c is associated to c.
Fix a ray in c1"no(c). Consider now a non-singular limit point b3S of the ray in c
1. Let
v03c with no(v0) very near b and in a foliated box containing b. Choose v
i3c with v
iescaping
the ray of c in question and no(vi)Pb. Let fibe the segment of c from v
0to v
i. The arc no(fi)
has both endpoints very near b and can be canonically closed up to a loop listarting at
no(v0). Then [li] represents an element g
iof n
1(S). The map g
itakes v
03c to a point which is
very near to vi3c in SI . Notice the v
iescape in c. Since c has a well-defined ideal point c*3S1
=,
it follows that gi(v
0)Pc* as iPR.
Up to subsequence assume that the sides of c and gi(c) containing singular free rays of
FI 6S
are the same. Then there are rays r(q) of FI 6S
starting in q3c and r(gi(q@)) starting in
gi(q@)3g
i(c), which share a subray and therefore their ideal points in S1
=are the same point
c3S1=
. The previous paragraph implies that the ideals point of r (gi(q@)) and r (g
i(q)) are the
same because both start in c and go to the singular free side. But r(gi(q)) and g
i(r(q)) are the
same ray, hence gi(r (q)) has ideal point c. Since r(q) and g
i(r(q)) have ideal point c, this
implies that gi(c)"c.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 857
Since n1(S) is a surface group, the stabilizer of c3S1
=is at most cyclic. Hence all g
iare
powers gni of a single covering translation g3n1(S). We may assume that all n
i'0, which
implies that niP#R when iPR. It follows that c is the attracting fixed point of g,
therefore gni(v0)Pc when iPR. But gni(v
0)"g
i(v
0)Pc* when iPR, therefore c*"c, that
is, c is the ideal point of the ray c.If no(c) is contained in a Reeb annulus or no(c) is closed the result follows immediately, so
suppose this is not the case. Therefore c is a quasigeodesic in oJ (SI ) with distinct ideal pointsin S1
=. Consider gj(c) with jP!R. One endpoint of gj(c) is c. The other endpoint of gj(c)
converges to the repelling fixed point of g. In addition, gj(c) are uniform K-quasigeodesics inthe hyperbolic plane and they are nested (that is gj(c) separates gj~1(c) from gj`1(c) forany j). Therefore, gj(c) converges to a leaf c@ of FI 4
Sand g(c@)"c@. Therefore no(c@) is a
closed leaf of F 4S. In addition, the fact that gn(c)Pc@ implies that no(c) has a ray which
spirals towards the closed leaf no(c@). This finishes the proof of Lemma 5.14. K
Remark. The situation in Lemmas 5.13 and 5.14 actually occurs quite often: supposethat F6
Shas a Reeb annulus A with a boundary component a. Topological considerations
imply that F4Shas a closed leaf b contained in the interior of A. Also we can choose b so that
nearby stable leaves spiral towards b. Lift a and b coherently to aJ , bI LSI . Then it is easy tosee that aJ and bI form a perfect fit. Notice that the rays of FI 6
Sstarting in aJ and in the side
containing bI are all contained in the lift of A to SI , hence they do not sweep out an entirecomponent of SI !aJ .
Acknowledgements—We thank Lee Mosher whose enthusiasm and conversations helped spur the results of thisarticle. We also thank Daryl Cooper for many interesting and useful suggestions.
REFERENCES
1. Anosov, D.V., Geodesic flows on closed Riemannian manifolds with negative curvature. Proceedings of theSteklov Institute of Mathematics, 1969, 90.
2. Barbot, T., Caracterisation des flots d’Anosov en dimension 3 par leurs feuilletages faibles. Ergodic ¹heoryDynamical Systems, 1995, 15, 247—270.
3. Bleiler, S. and Casson, A., Automorphims of surfaces after Nielsen and Thurston. London MathematicalSociety Student Texts, 9, Cambridge University Press, Cambridge, 1988.
4. Bonahon, F., Bouts des varietes hyperboliques. Annals of Mathematics, 1986, 124, 71—158.5. Brittenham, M., Essential laminations in Seifert fibered spaces. ¹opology, 1993, 32, 61—85.6. Cannon, J. and Thurston, W., Group invariant peano curves, to appear.7. Cooper, D., Long, D. and Reid, A., Bundles and finite foliations. Inventiones Mathematicae, 1994, 118,
255—283.8. Cooper, D., Long, D. and Reid, A., Finite foliations and similarity interval exchange maps. ¹opology, 1997, 36,
209—227.9. Fenley, S., Asymptotic properties of depth one foliations in hyperbolic 3-manifolds. Journal of Differential
Geometry, 1992, 36, 269—313.10. Fenley, S., Anosov flows in 3-manifolds. Annals of Mathematics, 1994, 139, 79—115.11. Fenley, S., One sided branching in Anosov foliations. Commentarii Mathematici Helvetici, 1995, 70, 248—266.12. Fenley, S., The structure of branching in Anosov flows of 3-manifolds. Commentarii Mathematici Helvetici,
1998, 73, 259—297.13. Fenley, S., Incompressible tori transverse to Anosov flows in 3-manifolds. Ergodic ¹heory Dynamical Systems,
1997, 17, 105—121.14. Fenley, S., Foliations with good geometry, to appear.15. Fenley, S. and Mosher, L., Quasigeodesic flows in hyperbolic 3-manifolds, to appear.16. Fried, D., Transitive Anosov flows and pseudo-Anosov flows. ¹opology, 1982, 21, 299—303.17. Gabai, D., Foliations and the topology of 3-manifolds. Journal of Differential Geometry, 1983, 18, 445—503.18. Gabai, D. and Oertel, U., Essential laminations and 3-manifolds. Annals of Mathematics, 1989, 130,
41—73.19. Greenberg, M. J. and Harper, J. R., Algebraic topology, a first course, Benjamin-Cummings, New York, 1981.20. Haefliger, A., Varietes feuilletees. Annali della Scuola Normale Superiore di Pisa, 1962, 16 (3), 367—397.
858 S. R. Fenley
21. Hector, G. and Hirsch, U., Introduction to the Geometry of Foliations, Part A, 2nd. edn. Vieweg, Braun-schweig, 1986.
22. Hempel, J., 3-Manifolds, Annals of Mathematics Studies 86, Princeton University Press, Princeton, NJ, 1976.23. Johannson, K., Homotopy Equivalences of 3-Manifolds With Boundaries, Lecture Notes in Mathematics 761,
Springer, Berlin, 1979.24. Levitt, G., Foliations and laminations in hyperbolic surfaces, ¹opology, 1983, 22, 119—135.25. Mangum, B., Thesis, University of California, Santa Barbara, 1996.26. Marden, A., The geometry of finitely generated Kleinian groups. Annals of Mathemtics, 1974, 99, 383—462.27. Mosher, L., Dynamical systems and the homology norm of a 3-manifold I. Efficient intersection of surfaces and
flows. Duke Mathematical Journal, 1992, 65, 449—500.28. Mosher, L., Dynamical systems and the homology norm of a 3-manifold II. Inventiones Mathematicae, 1992,
107, 243—281.29. Mosher, L., Laminations and flows transverse to finite depth foliations, preprint.30. Novikov, S. P., Topology of foliations. ¹ransactions of the Moscow Mathematical Society, 1963, 14, 268—305.31. Plante, J., Anosov flows, transversely affine foliations and a conjecture of Verjovsky. Journal of ¸ondon
Mathematical Society. Second Series, 1981, 23, 359—362.32. Reeb, G., Sur certaines proprietes topologiques des varietes feuilletees. Actualites Scientifiques Industrielles, No
1182, Herman, Paris, 1982.33. Stallings, J., On fibering certain 3-manifolds. ¹opology of 3-manifolds, Prentice-Hall, Englewood Cliffs, NJ,
1962, pp. 95—100.34. Strebel, K., Quadratic differentials. Springer, Berlin, 1984.35. Thurston, W., The Geometry and Topology of 3-Manifolds. Princeton University Lecture Notes, Princeton,
NJ, 1982.36. Thurston, W., Hyperbolic structures on 3-manifolds II : Surface groups and 3-manifolds that fiber over the
circle, preprint.37. Thurston, W., On the geometry and dynamics of diffeomorphisms of surfaces. Bulletin of the American
Mathematical Society, 1988, 19, 417—431.38. Thurston, W., 3-dimensional manifolds, Kleinian Groups and hyperbolic geometry. Bulletin of the American
Mathematical Society, 1982, 6 (new series), 357—381.39. Waldhausen, F., On irreducible 3-manifolds which are sufficiently large. Annals of Mathematics, 1968, 87,
56—88.40. Zeghib, A., Sur les feuilletages geodesiques continus des varietes hyperboliques. Inventiones Mathematicae,
1993, 114, 193—206.
Department of MathematicsPrinceton University and Washington UniversityFine Hall - Washington RoadPrinceton, NJ 08544-1000U.S.A.
SURFACES TRANSVERSE TO PSEUDO-ANOSOV FLOWS 859
Top Related